Due to the inherent structural complexity, dynamic behavior, and strong nonlinearities of interconnected systems, conventional control and existing data-driven methods such as T–S fuzzy models and SINDy often struggle to simultaneously ensure modeling accuracy, structural interpretability, and real-time adaptability under dynamic operation modes. To address these limitations, this paper proposes an affine-structured performance-controllable neural network guided backstepping predictive control framework. Unlike conventional black-box data-driven models, the proposed approach embeds the affine nonlinear system structure into the neural network and adopts a Lyapunov-based training strategy, enabling both accurate dynamic approximation and explicit control law design. Meanwhile, a backstepping predictive control scheme is developed to effectively handle interconnection-induced coupling and multivariable constraints with reduced computational burden. Furthermore, a performance-controllable integrated neural network adaptive update method is introduced by reformulating model adaptation as a control problem, allowing near-real-time model updating using only finite data and guaranteeing stable and rapid convergence of prediction errors. Rigorous theoretical analysis and extensive experimental results demonstrate that the proposed method achieves superior control performance under dynamic operation modes.

Industrial processes often exhibit multiple operational modes, driven by variations in production conditions, raw material properties, or operational settings. A single predictive model struggles to accommodate multimodal dynamics, leading to suboptimal performance in model predictive control (MPC). In addition, most multimode MPC approaches necessitate the development of separate predictive models for each operational mode, making the effectiveness of MPC heavily reliant on the mode switching strategy. In this paper, a novel adaptive mode-switching nonlinear predictive control (AMSNPC) method is proposed to improve both modeling robustness and control accuracy in dynamic industrial scenarios. First, a multimode modeling method based on continual learning framework is investigated to eliminate the need for explicit mode switching logic. Then, an adaptive error-triggered correction mechanism is designed to automatically detects mode switching based on output prediction errors and accelerate the response speed to the target values during operational mode switching. Finally, a heuristic optimization algorithm named state transition algorithm (STA) is adopted to find the global optimal control solution for nonlinear MPC problem. A numerical simulation experiment and an industrial case study are conducted to demonstrate that the AMSNPC method achieves high control accuracy and minimizes overshoot in multimode processes control.

The application of autonomous systems plays a crucial role in ensuring the safe and stable operation of industrial processes. Control methods grounded in artificial intelligence (AI) present new prospects for this application. However, the implementation of AI-based control methods typically depends on complete observation data. When unobservable states exist, it becomes challenging to achieve autonomous control. On the other hand, the process generally operates under dynamic working conditions, and the model needs to be updated in time to avoid model mismatch, which further increases the difficulty of autonomous control. To address these challenges, an autonomous control method based on a joint spatial-temporal model (AC-JSTM) is proposed to achieve autonomous control of unobservable point states of industrial processes under dynamic working conditions. Specifically, to solve the problem of model mismatch and poor control effect caused by the dynamic conditions of industrial processes, a condition identifier based on orthogonal test design (CI-OTD) is first proposed. The data set is constructed through typical condition parameter design, and the conditions are distinguished based on the correspondence between data distribution and condition labels. Then, considering the spatial distribution characteristics between unobservable and observable points and the dynamic correlation of observable points in the time dimension, a JSTM is established. The features of spatial and dynamic dimensions are integrated by a joint training method to achieve the prediction of unobservable points based on observable points. Finally, combining CI-OTD and JSTM, a dynamic working condition control (DWCC) framework is established to achieve autonomous control of unobservable points under dynamic working conditions. To verify the superiority and effectiveness of the proposed method, control experiments are designed for both catalytic rods and tubular reactors. The results show that the proposed method can achieve accurate control of unobservable points under dynamic conditions, and the control accuracy is improved by 18.69% compared with the control framework based on the spatial-temporal model.

Image data serve as a vital information carrier in industrial systems, but closed-set recognition methods falter in dynamic online environments with continually emerging classes. Online learning adapts rapidly but suffers from catastrophic forgetting, while continual learning alleviates forgetting through regularization yet depends on offline training, limiting adaptability to rapid task shifts. Resource-constrained edge devices demand real-time model updates, where streaming few-shot novel classes introduce feature bias, reducing class separability and accelerating model degradation. To this end, this work proposes the Biscale Storage and bilevel contrast-based continual feature evolution learning (Bi2CFEL) for industrial online recognition. By integrating biscale storage and bilevel contrast, Bi2CFEL efficiently memorizes and captures intraclass and interclass information, enabling accurate and robust image recognition in industrial online scenarios. First, spatio-temporal biscale guided storage strategy is innovatively proposed, constructing a representative and discriminative historical storage through class prototype distance screening criterion and prototype calibration mechanism. Second, to address overfitting risks in limited storage, distribution-aware memory reactivation method reactivates old-class memory based on stored sample distributions, effectively suppressing feature drift and forgetting. Finally, bilevel contrastive dictionary learning framework optimizes high-dimensional sparse feature spaces via instance-prototype cross-level contrast mechanisms, enhancing intraclass cohesion and interclass separation to decouple old and new class features. Experimental results demonstrate that Bi2CFEL achieves state-of-the-art performance in generalization capability, forgetting rate, and overall process performance.

Modern industrial processes are highly complex and dynamically evolving, with new modes emerging due to variations in raw materials, production environments, and other factors. Traditional monitoring methods often suffer from catastrophic forgetting during updates, where old knowledge is overridden. Although continual learning offers an effective solution, existing methods tend to overprotect historical knowledge, limiting the adaptability to new modes. To address the above problems, we propose orthogonal decoupled continual dictionary learning (ODCDL), which orthogonally decouples the dictionary space into stability and plasticity subspaces. The stability space preserves representations of historical modes to mitigate forgetting, while the plasticity space allows flexible updates for learning new modes. To balance the tradeoff between stability and plasticity, we impose dynamic constraints with varying strengths on the two subspaces. This design enhances both retention of old knowledge and learning of new knowledge, ensuring accurate monitoring across all operating conditions. Extensive experiments have demonstrated that the proposed method outperforms several state-of-the-art methods, exhibiting superior continual learning capability and process monitoring performance.

Focusing on event-triggered robust state estimation, this study examines nonlinear networked systems under denial-of-service attacks and limited communication rates. Linearization errors arising in the extended Kalman filter increase state-estimation error, and denial-of-service–induced network overloads cause packet dropouts that disrupt remote robust estimation and degrade measurement delivery. To alleviate these effects, we develop an event-triggered robust state estimator and establish, under standard observability, packet-arrival, and trigger-threshold conditions, the uniform mean-square boundedness of the estimation error, with explicit upper bounds and an admissible design range. Numerical simulations demonstrate that the proposed estimator markedly improves estimation accuracy.

Graph contrastive learning (GCL) is a powerful self-supervised learning approach. However, existing GCL methods are designed for homophilic graphs, using low-pass filters that struggle to capture high-frequency components in heterophilic graphs. We proposeGraphContrastiveLearning withRegularization and stabilization techniques enhanced high-passFilter (GCLRF).REgularization andStabilization techniques enhancedHigh-pass filter (RESH) can serve as a mutually promoting plug-in, significantly improving the performance of various homophilic GCL training strategies on heterophilic graphs. We also investigate four component orderings in RESH and identify the optimal fusion mechanism, demonstrating its critical impact on performance. Experiments show GCLRF achieves state-of-the-art (SOTA) performance across six benchmark datasets in node classification and clustering. Notably, on the Cornell dataset, GCLRF outperformers classification accuracy by 6.76% and achieves a 23.64%relative improvement in clustering normalized mutual information (NMI).

With the advancing application of Unmanned Aerial Vehicles (UAVs, also known as Drones), UAV swarms are deployed in various complex mission scenarios to collaborate on particular assignments. However, UAV group communication encounters security and privacy challenges due to the open and insecure communication environment. Moreover, many existing UAV group authentication and key agreement (GAKA) schemes provide limited support for dynamic membership and fail to enforce mandatory participation of the leader UAV in hierarchical architectures. To address these limitations, we propose a blockchain-enabled scalable GAKA scheme tailored for dynamic IoD, named BSGAKA-IoD. The scheme achieves UAV threshold GAKA through an enhanced designated participant$ (ID, t, n)^*$-secret sharing scheme, which cryptographically enforces leader participation and enables threshold-based GAKA among the cluster leader and member UAVs. The scheme supports efficient group authentication, secure group key establishment, and dynamic joining/leaving without requiring any secure channel, while achieving$ O(n)$computational and communication complexity. We conduct comprehensive security analysis and develop two blockchain prototypes: a Solidity smart contract implementation validated in Remix, and a Hyperledger Fabric implementation benchmarked using Hyperledger Caliper under two network configurations. The results confirm that BSGAKA-IoD provides strong security guarantees, maintains linear scalability, and is feasible for deployment in realistic consortium-chain IoD settings.

Modern industrial systems typically operate in a multimode environment due to continuous changes in raw materials and production schedules. Traditional static monitoring methods often struggle to accurately capture the intricate relationships between different modes, leading to model mismatches and a significant decline in monitoring performance. On the other hand, emergence of a new mode often prompts model updates that focus on the new data, potentially causing the model to forget historical knowledge, which is a phenomenon known as catastrophic forgetting. To address these issues, we propose a zero forgetting lifelong dictionary learning (ZFLDL) method for multimode process monitoring. ZFLDL integrates a low-rank basis matrix growth mechanism with an adaptive basis matrix selector, enabling effective representation of knowledge across both new and historical modes. Specifically, a novel low-rank decomposition method is first proposed to address feature redundancy in data representation with an overcomplete dictionary. This method utilizes a low-rank matrix, i.e. a “basis matrix”, as the fundamental representation unit, which can accurately capture the core data features and enhance the dictionary's representation capability. Then, to address the emergence of new modes, a zero forgetting continual learning framework based on low-rank basis matrix growth is proposed. By adaptively adding basis matrices for each new mode while freezing those associated with historical modes, the dictionary efficiently assimilates new knowledge while ensuring zero forgetting of historical knowledge. Additionally, since only the basis matrices related to the new mode are updated, this approach significantly reduces the number of parameters requiring training, thus enhancing the model's learning efficiency. Finally, an adaptive basis matrix selector is proposed. By evaluating the contribution of each basis matrix to different modes, it dynamically selects the basis matrices that best match the specific mode, ensuring accurate mode recognition and reliable process monitoring. Extensive experiments have demonstrated that the proposed method outperforms several state-of-the-art methods, exhibiting superior continual learning capability and enhanced process monitoring performance.

Recently, large language models (LLMs) have demonstrated remarkable capabilities in understanding and generating natural language content, attracting widespread attention in both industry and academia. An increasing number of services offer LLMs for various tasks via APIs. Different LLMs demonstrate expertise in different domains of queries (e.g., text classification queries). Meanwhile, LLMs of different scales, complexities, and performance are priced diversely. Driven by this, several researchers are investigating strategies for selecting an ensemble of LLMs, aiming to decrease overall usage costs while enhancing performance. However, to our best knowledge, none of the existing works addresses the problem, how to find an LLM ensemble subject to a cost budget, which maximizes the ensemble performance with guarantees. In this paper, we formalize the performance of an ensemble of models (LLMs) using the notion of correctness probability, which we formally define. We develop an approach for aggregating responses from multiple LLMs to enhance ensemble performance. Building on this, we formulate the Optimal Ensemble Selection (OES) problem of selecting a set of LLMs subject to a cost budget that maximizes the overall correctness probability. We show that the correctness probability function is non-decreasing and non-submodular and provide evidence that the OES problem is likely to be NP-hard. By leveraging a submodular function that upper bounds correctness probability, we develop an algorithm, ThriftLLM, and prove that it achieves an instance-dependent approximation guarantee with high probability. Our framework functions as a data processing system that selects appropriate LLM operators to deliver high-quality results under budget constraints. It achieves state-of-the-art performance for text classification and entity matching queries on multiple real-world datasets against various baselines in our extensive experimental evaluation, while using a relatively lower cost budget, strongly supporting the effectiveness and superiority of our method.

In response to the challenge of strongly nonlinear and multimode systems control, this paper introduces a weighted deep learning based adaptive predictive control method. This approach integrates LSTM networks for different operating modes using a set of weighting coefficients. These coefficients are dynamically updated during online control via an error-guided scheduling strategy to adapt to changing operation modes. Compared to offline identification based methods, the proposed method eliminates the need for mode recognition or model switching strategies and can adapt to drifted operation modes. In contrast to online methods, it achieves rapid model convergence and reduced computational cost, requiring only minimal data to update the weighting coefficients without necessitating the retraining of the LSTM networks. Theoretical convergence and stability analysis ensure the reliability of the proposed method. Numerical simulations and industrial control experiments demonstrate that the proposed approach exhibits favorable control performance across both known and drifted operation modes. Note to Practitioners—Considering the changing operation modes in complex industrial processes and the detrimental effect of slow or unstable control during system operation, this paper proposes a weighted LSTM based predictive control method for strongly nonlinear and multimode systems. Extensive experiments demonstrate that compared to other state-of-the-art methods, this method can rapidly adapt to changes in operating modes with a small amount of data, meeting both real-time and stability requirements for online control.

Modern industrial processes typically operate under variable conditions due to fluctuations in raw materials and production loads. Consequently, accurate monitoring across multiple operational modes is essential for process optimization. Although some global modeling based methods have been proposed, they struggle to adapt to the continual emergence of new modes. On the other hand, online updating monitoring methods often suffer from “catastrophic forgetting”, i.e., they lose the ability to represent historical modes. To address these challenges, this article proposes a multimode process monitoring method based on Incremental Rank Continual Dictionary Learning (IRCDL). The method integrates the incremental learning framework based on low-rank matrix augmentation with the adaptive weight selector, allowing the dictionary to retain the knowledge of both historical and new modes. Specifically, to address the issue of information redundancy in overcomplete dictionaries, an innovative dictionary construction method based on low-rank matrices weighting is proposed. Using low-rank matrices as the fundamental units for dictionary construction, this method accurately captures the intrinsic features of different modes while significantly reducing redundant parameters. Then, to handle the continual emergence of new modes, an incremental learning framework based on low-rank matrix augmentation is developed. This framework continually adds new low-rank matrices to learn representations of new modes, while keeping the historical low-rank matrices intact to maintain “zero-forgetting” of historical modes, thereby enabling accurate monitoring of multimode processes. Finally, an adaptive weight selector is designed to automatically optimize the weights of low-rank matrices, thereby enhancing the representation of different modes and significantly improving the performance of multimode monitoring. Extensive experiments have demonstrated that the proposed method achieves state-of-the-art continual learning capabilities (i.e., “zero forgetting”) and exceptional process monitoring performance.

Industrial systems often undergo dynamic changes during operation, which presents challenges for traditional identification and control methods. These challenges arise in two aspects: variations in model structure and parameters, and differences in control objectives across diverse operating conditions. Traditional static predictive control methods face challenges in meeting the high-precision, real-time requirements in practice. In addition, control schemes with fixed parameters encounter difficulties in adapting to varying control objectives, resulting in suboptimal control performance. To address these problems, this article proposes a dynamic error-triggered adaptive control method, which can identify the operating conditions and objectives in real-time. Specifically, a dynamic error-triggered model updating mechanism is first established to detect changes in operating conditions and update the prediction model. To overcome the model mismatch during the transition process, a novel enhanced transition control (ETC) method is proposed, which designs a transition error corrector to decline prediction error and a high informative pseudo-random binary sequence (HIPRBS) input to enhance the excitation level. Considering the differences in control objectives under varying operating conditions, a fuzzy weight-adaptive method is proposed to balance heterogeneous indicators in different conditions. Two types of systems, high-speed and high-stability, are designed to validate the superiority of the proposed method. Extensive experimental results demonstrate that, compared to some state-of-the-art methods, the proposed method can efficiently and accurately identify emerging operating conditions, dynamically adjust optimization objectives, and achieve real-time control effects under varying operation conditions.Note to Practitioners—The motivation of this paper is to develop a high-precision and real-time control method for industrial systems that operate under frequently changing conditions. The proposed method can adapt to changes in model parameters and control objectives in multiple operating conditions processes. Compared with some state-of-the-art methods, this method significantly enhances the control performance in the transition process of mode switching, meeting the long-term stable operation requirements of industrial systems.

Modern industrial systems possess the capacity to accumulate substantial data, thereby enabling data-driven process monitoring. However, the dynamic industrial processes give rise to new working conditions continuously. Traditional methods require substantial new data for updates, resulting in considerable delays. Furthermore, edge devices face resource limitations, which complicates the ability to meet the increasing demands for storage and computing power. On the other hand, the frequent transmission of data between the cloud and edge introduces potential security risks. To address these challenges, this paper proposes a Secure Monitoring Method based on Self-Updating Dictionary Learning (SUDL-SM) within a cloud-edge collaboration framework. Specifically, to tackle the issue of poor model adaptability caused by limited data of new modes, this paper first proposes a dictionary learning method based on multi-task hardness evaluation. By evaluating the multi-dimensional contributions of samples, those with strong generalization are extracted, and the dictionary is updated online accordingly, ensuring adaptability to both new and historical conditions. Subsequently, due to the resource constraints inherent in edge devices, a dictionary distillation compression method has been proposed. This method aims to maximize dictionary compression while preserving the original monitoring performance, thereby ensuring efficient and accurate inference on edge devices. Finally, a hybrid encryption-based cloud and edge data transmission protocol is designed to effectively address malicious activities such as data theft and tampering by ensuring reliable interaction between the cloud and the edge. Extensive experiments verified the effectiveness and superiority of the proposed method.

During the operation of a distributed parameter system (DPS), its working conditions typically undergo dynamic changes. Although online learning methods can enable models to adapt to new working conditions to some extent, they often confront the “catastrophic forgetting” problem, where the updated model forgets historical working conditions. On the other hand, only a few new samples can be collected during online operation, and the sparse samples make it difficult to establish accurate models for new working conditions. Therefore, achieving precise control under full working conditions remains a challenging problem. To address these challenges, this paper proposes an adaptive predictive control method based on continuous learning that achieves stable control under full working conditions by continuously identifying working conditions and triggering adaptive model updates in real time. Specifically, a spatial-temporal feature-based working condition identification method is first proposed to identify changes in working conditions automatically. Then, to address the challenge of limited data for model updating, a parameter transfer method is proposed. Simultaneously, to ensure that the updated model retains the ability to characterize historical working conditions, an Elastic Weight Consolidation (EWC) constraint is incorporated into the loss function, thus overcoming the catastrophic forgetting problem, and ensuring the updated model can represent both the historical and new working conditions. Finally, by incorporating this condition identification mechanism and adaptive predictive model into the model predictive control framework, continuous precise control of DPS can be achieved. To demonstrate the superiority of the proposed method, extensive experiments are designed. Experimental results show that the proposed method can accurately identify new working conditions and learn new condition models with only a few online samples while overcoming model mismatch of historical conditions, ultimately achieving high-precision control under full working conditions.Note to Practitioners—Motivated by the fact that DPS often operates in different conditions and online learning methods confront “catastrophic forgetting” and “sparse sample training” problems, this paper proposes an adaptive learning control method. The proposed method can learn new working condition features with small online samples while retaining the ability to represent historical working conditions, and thus achieves accurate control under full working conditions.

Complex systems are frequently influenced by uncertain factors, making it difficult for traditional fixed-model control schemes to achieve high-precision control. Data-driven control methods offer a solution, but they face challenges in constructing accurate models due to insufficient excitation in operational data. Moreover, mismatches between historical models and new conditions coupled with limited data accumulation under new conditions reduces the operational performance throughout the entire process. To address these issues, this paper proposes a generalized integrated fuzzy model predictive control (GIF-MPC) framework. It combines the generalization capability of fuzzy control with the precision of model predictive control to ensure highprecision control under all conditions. Specifically, a strategy switching mechanism, triggered by a mismatch characteristic parameter is first proposed, which transitions the original strategy to a fuzzy-driven excitation control method, thereby mitigating the control performance degradation caused by the mismatch between control strategies and complex systems. Then, a fuzzy control feature extraction method is proposed to balance fuzzy set activation and improve adaptability to unknown conditions. Additionally, an optimal input excitation design method is proposed to tackle insufficient data excitation, enabling effective control. Once sufficient data is accumulated, the model switches to model predictive control. The dual decision mechanism guided by the data information and triggered by the mismatch characteristic parameter effectively ensures high precision control under uncertainties. Numerical experiments demonstrate that the GIF-MPC method ensures high-precision control throughout disturbances and condition changes. The solution is also successfully deployed in an industrial setting, validating its excellent control performance under full operation conditions.

Maintaining an optimal alumina concentration distribution is crucial for ensuring stable operation and reducing energy consumption in aluminum electrolysis cells. Due to the strong coupling of multiple physical fields within the electrolysis cell, the rapid and accurate prediction of alumina concentration distributions has long been a formidable challenge. Furthermore, the harsh industrial environment only allows for the acquisition of alumina concentration data at limited locations, hindering the construction of precise predictive models for the alumina concentration distribution. Moreover, the presence of measurable disturbances, such as feeding, can exert additional influences on the alumina concentration, necessitating traditional methods to collect new data and reconstruct models. To address these issues, this article proposes a structural process knowledge-informed neural network for alumina concentration distribution prediction with desirable precision and high solving efficiency. Specifically, considering the alumina concentration distribution is primarily driven by electrolyte convection and secondarily by electrolysis consumption, pretraining modules for flow and concentration field are designed to explicitly modify the network structure, thereby strengthening the relationship between inputs and process state variables, which significantly enhance prediction efficiency and accuracy. Subsequently, a loss function incorporating process governing equations is designed, which effectively constrains the relationship between inputs and process state variables to adhere to physical principles. This enables accurate concentration prediction at any location in the field, even with a limited number of spatial training data. Finally, a training method for alumina concentration prediction under varying initial concentrations is introduced, incorporating the initial concentration and a fixed time step into the network inputs, thereby eliminating the need for traditional methods to reconstruct models for continuous predictions under different initial conditions. Extensive experiments demonstrate that our method achieves significant performance improvements compared to some state-of-the-art methods.

With the rapid development of modern industry and the increasing prominence of artificial intelligence, data-driven process monitoring methods have gained significant popularity in industrial systems. Traditional static monitoring models struggle to represent the new modes that arise in industrial production processes due to changes in production environments and operating conditions. Retraining these models to address the changes often leads to high computational complexity. To address this issue, we propose a multimode process monitoring method based on element-aware lifelong dictionary learning (EaLDL). This method initially treats dictionary elements as fundamental units and measures the global importance of dictionary elements from the perspective of the multimode global learning process. Subsequently, to ensure that the dictionary can represent new modes without losing the representation capability of historical modes during the updating process, we construct a novel surrogate loss to impose constraints on the update of dictionary elements. This constraint enables the continuous updating of the dictionary learning (DL) method to accommodate new modes without compromising the representation of previous modes. Finally, to evaluate the effectiveness of the proposed method, we perform comprehensive experiments on numerical simulations as well as an industrial process. A comparison is made with several advanced process monitoring methods to assess its performance. Experimental results demonstrate that our proposed method achieves a favorable balance between learning new modes and retaining the memory of historical modes. Moreover, the proposed method exhibits insensitivity to initial points, delivering satisfactory results under various initial conditions.

Multirate sampling data are common in industrial systems, and traditional diagnostic models face challenges in directly handling such data due to its incomplete characteristics. Recently, some data reorganization methods have been proposed, which divide the multirate sampling data into several data subsets and learn diagnostic models based on each subset separately. However, they always treat different tasks independently and ignore the correlation between data subsets. In addition, training individual models for each task will decrease the data available and increase computational resources. In this article, a novel attention-based mask network fault diagnosis model for multirate sampling data is proposed, which converts the multirate sampling fault diagnosis problem into a continual multitask learning problem and realizes it with one model. Moreover, an attention mechanism and cumulative gradients are integrated during multitask learning to alleviate the catastrophic forgetting problem, thus further guaranteeing the performance of fault diagnosis. Notably, the proposed method is the first to approach multirate sampling data from a multitask continual learning perspective, allowing it to fully leverage raw data. Extensive experiments, including a numerical simulation and an industrial three-phase flow case, indicate the proposed method effectively tackles multirate sampling data fault diagnosis within a multitask continual learning framework. Since, it takes the correlation between such data and avoids the detrimental effects of catastrophic forgetting, it achieves higher accuracy in fault diagnosis.

Complex industrial processes often run under varying operating conditions. Learning-based control methods are difficult to adapt to these unknown variations. Therefore, it is necessary to update the model and control strategy adaptively. However, in traditional control frameworks, due to the limitation of computational and storage resources of edge devices, control strategies are difficult to update once deployed, which leads to model mismatch after operating condition change and seriously reduces the control performance. To solve this problem, this article proposes a novel cloud-edge collaborative control method. Specifically, a cloud-assisted parallel subspace identification method is proposed, which fully utilizes the powerful computational capability of the distributed cluster in the cloud to achieve fast and accurate model identification. Then, an explicit control strategy is proposed, which solves the control law as a piece-wise affine function offline. The process model and explicit control law are sent down to the edge, enabling fast and precise control under limited resource constraints. An operating condition change detection method based on the process model is proposed, and the edge detects the emergence of new operating conditions by the prediction error. Meanwhile, to fully excite new operating condition characteristics, a joint control and excitation signal generator (JCESG) is designed. JCESG ensures accurate identification of new operating condition model under limited data, which in turn greatly shortens the operation condition switching process and ensures fast modeling and precise control in new operating conditions. Notably, considering that the proposed method can adaptively realize model identification and control law update, it is capable of adapting to the continuous change of operating conditions, and the sufficient excitation of JCESG greatly reduces the data volume requirement for model update, which further ensures that the method adapts to the full range of operating conditions. Finally, extensive experiments verified the superiority of the proposed method.

Spectral Graph Neural Networks (GNNs), alternatively known as graph filters, have gained increasing prevalence for heterophily graphs. Optimal graph filters rely on Laplacian eigendecomposition for Fourier transform. In an attempt to avert prohibitive computations, numerous polynomial filters have been proposed. However, polynomials in the majority of these filters are predefined and remain fixed across different graphs, failing to accommodate the varying degrees of heterophily. Addressing this gap, we demystify the intrinsic correlation between the spectral property of desired polynomial bases and the heterophily degrees via thorough theoretical analyses. Subsequently, we develop a novel adaptive heterophily basis wherein the basis vectors mutually form angles reflecting the heterophily degree of the graph. We integrate this heterophily basis with the homophily basis to construct a universal polynomial basis UniBasis, which devises a polynomial filter based graph neural network – UniFilter. It optimizes the convolution and propagation in GNN, thus effectively limiting over-smoothing and alleviating over-squashing. Our extensive experiments, conducted on datasets with a diverse range of heterophily, support the superiority of UniBasis in the universality but also its proficiency in graph explanation.

Graph Neural Networks (GNNs), known as spectral graph filters, find a wide range of applications in web networks. To bypass eigendecomposition, polynomial graph filters are proposed to approximate graph filters by leveraging various polynomial bases for filter training. However, no existing studies have explored the diverse polynomial graph filters from a unified perspective for optimization.

The study of continuous-time information diffusion has been an important area of research for many applications in recent years. When only the diffusion traces (cascades) are accessible, cascade-based network inference and influence estimation are two essential problems to explore. Alas, existing methods exhibit limited capability to infer and process networks with more than a few thousand nodes, suffering from scalability issues. In this paper, we view the diffusion process as a continuous-time dynamical system, based on which we establish a continuous-time diffusion model. Subsequently, we instantiate the model to a scalable and effective framework (FIM) to approximate the diffusion propagation from available cascades, thereby inferring the underlying network structure. Furthermore, we undertake an analysis of the approximation error of FIM for network inference. To achieve the desired scalability for influence estimation, we devise an advanced sampling technique and significantly boost the efficiency. We also quantify the effect of the approximation error on influence estimation theoretically. Experimental results showcase the effectiveness and superior scalability of FIM on network inference and influence estimation.

Unmanned aerial vehicles (UAVs, also known as Drones) have been widely employed in military defense and civilian service. However, as UAVs communicate over insecure open wireless channels, the security challenges and privacy concerns are becoming increasingly prominent. Moreover, some existing schemes to achieve authentication and key agreement (AKA) among UAVs are spliced with the assistance of two UAV-2-GCS mechanisms, which are not flexible enough to be applied in the Internet of Drones (IoD) scenarios. This article proposes a blockchain-assisted AKA scheme for UAVs networks (BAKAS-UAV) referred to as BAKAS-UAV, which addresses security and privacy concerns and overcomes high computational and communication costs in the IoD. A blockchain-based network model is presented in which the ground station acts as an edge node and manages the blockchain, which assists AKA. Based on the network model, both types of AKA mechanisms, UAV-2-GCS and UAV-2-UAV, are proposed, respectively. In particular, the ground control station (GCS) does not participate in the AKA of UAV-2-UAV process; only upon the process is completed the two UAVs synchronize the updated information with GCS. We also implement a smart contract as the authentication service, and the experimental implementation demonstrates the availability of our scheme in IoD. Physical unclonable functions (PUFs) is introduced on the UAVs side to defend against physical capture attacks and also to implement AKA mechanisms. The semantic security is proved formally based on the real-or-random (ROR) model, and the informal analysis shows that the scheme satisfies the demanded security requirements. The scheme’s performance is evaluated by simulating the UAVs and GCS settings with Raspberry Pi 4B and MacOS platforms, respectively, with implementation of several cryptographic primitives. The experimental results show that BAKAS-UAV achieves high efficiency.

With the development of information technologies, a large number of sensors have been deployed to obtain industrial data. As a result, data-driven approaches have become a crucial means for modeling of distributed parameter systems (DPS). However, due to harsh environments and unreliable sensors, data is often of low quality in practice, which in turn poses a challenge for data-driven modeling approaches. In order to address the challenge of inaccurate modeling of DPS induced by outliers, this paper proposes a physical-informed sparse learning method to overcome the adverse effects of outliers and achieve robust modeling of DPS by fully exploring the spatiotemporal dynamic of DPS. Specifically, this paper proposes an innovative method for robust modeling of DPS. The method incorporates the statistical features of contaminated data to restore the dynamic evolution structure of DPS, which weakens the adverse effects of outliers and address the problem of low modeling accuracy caused by contaminated observation data. Furthermore, the underlying partial differential equation (PDE) of DPS is incorporated into the constraint on the temporal deviation data, which leads to a physical-informed optimization objective and improves the reliability of outlier extraction and DPS modeling. Finally, an optimization algorithm based on the alternating direction method of multipliers (ADMM) with an adaptive penalty factor is proposed. This ensures the convergence of multivariate optimization problem and superior performance of the DPS modeling framework. Extensive experimental results have verified that the proposed method is effective in overcoming the adverse effects of outliers and achieving robust modeling of DPS.Note to Practitioners—The motivation of this paper is to develop an interpretable and robust modeling method for DPS. Considering the negative impact of outliers, the proposed method first restores the dynamic properties of the data based on the characteristics of the outliers. Then, the embedding of physical knowledge ensures the reliability of outlier removal and robustness of modeling. Extensive experimental results have verified that the proposed method can effectively overcome the adverse effects of outliers and outperform some state-of-the-art methods. Therefore, it is more suitable for real industrial systems.

Non-ferrous metals, as important basic raw materials, are the strategic supports for national economic development. For non-ferrous metal smelting enterprises, raw material procurement is the focal and most important session. Due to the fluctuation of production volumes and the future changes in raw-material prices, the procurement cost of raw materials is high and with a high risk of shortage. In this paper, we propose a multi-period rolling robust procurement model considering price and demand uncertainties. In particular, we design a data-driven method to construct the budget-based uncertainty sets and derive the robust counterpart of the robust procurement model. Comparative experiments on the real data with classic and advanced procurement policies show that our proposed solution approach achieves the lowest cost under the premise of continuous supply of raw materials. Interestingly, we observe that limited capital and warehouse capacity can effectively restrain unreasonable behavior and thus not to cause big losses in uncertain environments. In addition, a relatively long planning horizon can be counterproductive. These valuable and actionable insights can well guide practical decision-making.Note to Practitioners—For the raw material procurement of non-ferrous metal smelter, this article proposes a multi-period rolling robust procurement model considering price and demand uncertainties. Taking account of the dynamic characteristics of raw-material prices and the seasonal characteristics of raw-material demands, a data-driven method to construct budget-based uncertainty sets is designed. In particular, we derive the solvable robust counterpart of the robust procurement model. The proposed approach can reduce costs ensuring the continuous supply of raw materials. Some interesting and actionable managerial insights are obtained that can well guide practical decision-making, and the proposed data-driven approach is realizable.

Nowadays, the intelligent fault diagnosis problem of modern industrial systems has received increasing attention. However, with the increasing scale of industrial systems, the same category of faulty data often contains multiple working conditions, which leads to the multi-granularity of faulty data and the increasing complexity of data distribution. The possible existence of fine-granularity inter-class similarities in multi-granularity data can interfere with coarse-granularity recognition tasks, which brings challenges to data-driven fault diagnosis tasks. To solve the above issue, a fault diagnosis method based on multi-granularity dictionary learning is proposed in this paper. Specifically, the proposed method integrates prior knowledge into the dictionary learning framework by introducing the concept of “centroid atom”, which embeds the multi-granularity structure in the dictionary and improves the discrimination of dictionary atoms. In order to validate the proposed approach, experiments are conducted on a numerical simulation case and a zinc smelting roasting process. Compared with some state-of-the-art methods, the proposed approach performs better in the classification accuracy which shows that it can make full use of the knowledge provided by the multi-granularity structure while it is difficult to be modeled by the other methods. Therefore, it can be applied to fault diagnosis tasks more precisely and efficiently. Note to Practitioners—Motivated by the fact that industrial faults often occur in different modes, and different faults may occur in the same mode, the fault data often has multi-granularity characteristics, this paper proposes a multi-granularity dictionary learning method for complex industrial fault diagnosis. The proposed method can learn the multi-granularity structure in the data, and the obtained model contains both coarse-grained and fine-grained information, which effectively improves the fault diagnosis accuracy. Intensive experimental results show that the proposed method performs better than some state-of-the-art methods, it can make full use of the knowledge provided by the multi-granularity structure, while it is difficult to be modeled by others. Above all, it is verified as suitable for fault diagnosis of real industrial systems.

In this paper, we investigate the remote robust state estimation problems for nonlinear cyber-physical systems under denial-of-service attacks. The well-known extended Kalman filter is a commonly used method for state estimation of nonlinear systems. However, it does not take into account unmodeled dynamics or parametric uncertainties caused by first-order approximations, which makes its estimation performance unsatisfactory. In addition, denial-of-service attacks prevent the sensor from sending measurements to a remote state estimator by congesting the communication channel, which further deteriorates the estimation performance. To surmount these problems, a robust state estimation algorithm is developed based on sensitivity penalization, taking into account an explicit packet loss parameter. Under certain conditions, the boundedness of estimation error is proved. By selecting a negative resistance oscillation circuit for numerical simulations, it is demonstrated that the remote robust state estimator can markedly improve the estimation performance.

The intelligent goal of process manufacturing is to achieve high efficiency and greening of the entire production. Whereas the information system it used is functionally independent, resulting to knowledge gaps between each level. Decision-making still requires lots of knowledge workers making manually. The industrial metaverse is a necessary means to bridge the knowledge gaps by sharing and collaborative decision-making. Considering the safety and stability requirements of the process manufacturing, this article conducts a thorough survey on the process manufacturing intelligence empowered by industrial metaverse. First, it analyzes the current status and challenges of process manufacturing intelligence, and then summarizes the latest developments about key enabling technologies of industrial metaverse, such as interconnection technologies, artificial intelligence, cloud-edge computing, digital twin (DT), immersive interaction, and blockchain technology. On this basis, taking into account the characteristics of process manufacturing, a construction approach and architecture for the process industrial metaverse is proposed: a virtual-real fused industrial metaverse construction method that combines DTs with physical avatar, which can effectively ensure the safety of metaverse's application in industrial scenarios. Finally, we conducted preliminary exploration and research, to prove the feasibility of proposed method.

In this study, we propose a dynamics-learning multirate estimation approach to perceive the quality-related indices (QRIs) of the feeding solution of a unit process. A quality-related index for estimation is an intermediate technical indicator between a unit process and a proceeding unit process; hence, the estimation problem is formulated as a two-stage estimation problem utilizing the production data of both unit processes. Dynamics-learning bidirectional long short-term memory (BiLSTM) with different inputs for the forward and backward layers is proposed to manage the input data from the different unit processes. In the dynamics-learning BiLSTM, a cycle control gate is added in the memory cell to learn the dynamics of the QRIs, thereby enabling a high-rate estimation under multirate conditions. A Bayesian estimation model is then combined with the dynamics-learning BiLSTM model to manage the process delay. Ablation and comparative experiments are conducted to evaluate the feasibility and effectiveness of the proposed estimation approach. The experimental results illustrate the performance and high-rate estimation ability of the proposed approach.

Modern industrial processes often exhibit complex and uncertain operating state fluctuations due to the diversification of production materials, the complexity of production processes, and the harsh production environment. To cope with the operating state fluctuations of industrial processes, fuzzy neural networks, as a form of human-inspired computing, have been introduced and widely applied. In contrast, model predictive control (MPC) typically employs a multi-model control strategy for the multimode process. However, ensuring control performance during the switching phase and meeting real-time control requirements poses challenges. To address the real-time control challenge in multimode processes, this article proposes a learning framework for fuzzy neural network explicit control for full operation conditions. The proposed framework includes two modules: the fuzzy neural network-based explicit control law (FNNECL) and the neural network-based predictive model (NNPM). By adjusting the structure and parameters of FNNECL, precise control of full operation conditions is achieved. Specifically, a novel truncated radial basis activation function is first presented, and the operation condition change measurement index of data coverage is proposed to identify the mismatch degree between current and historical operation conditions accurately. Then, for small-scale operation condition changes, an elastic weight consolidation (EWC) mechanism is introduced to ensure that FNNECL can learn control strategies for new operation conditions while maintaining control performance for historical operation conditions. Finally, to address large-scale operation condition changes, a radial basis function (RBF) neuron growth mechanism based on data coverage is proposed. This mechanism calculates the data coverage of fuzzy rules and selectively adds new neurons to the fixed structure FNNECL. This enables the explicit control law to fit large-scale changes and effectively learn control strategies for new operation conditions. It is worth noting that during the application of this method, there is no need to know the current operation conditions in advance. Based on a single FNNECL model, accurate control sequences can be rapidly obtained, overcoming the traditional online optimization problem of multiple models. This greatly expands the application scope and achieves precise control under full operation conditions. Extensive experiments including numerical simulation and industrial roasting process verified the feasibility and effectiveness of the proposed method.

Deep neural networks (DNNs) as one of the key enabling technologies have been widely used in industrial artificial intelligence (IAI). However, recent research has revealed that they are quite vulnerable to adversarial attacks, arousing serious concerns about DNNs' robustness in many IAI-driven applications such as industrial video analysis tasks. Considering the attack efficiency and effectiveness, it is essential to study the sparse adversarial attack examples. Nevertheless, current methods' performance is limited by insufficient sparsity and lacks a unified framework. To solve these problems, in this article, we focus on sparse adversarial video attacks and propose a dual-branch neural network-based model to generate sparse adversarial video examples in an end-to-end fashion. We conduct extensive experiments with mainstream video models on public datasets and industrial case. Experimental results demonstrate that compared with state-of-the-art methods, our method can achieve a faster and better attacking performance with less than 1% perturbed pixels in the video.

Zinc rotary kiln is an important equipment in the nonferrous metallurgical industry. Due to unclear internal working conditions, its operation based on experience is random. Digital twin (DT) with virtual–real integration and synchronization ability is a necessary method to realize real-time and accurate monitoring for key variables, while its nowadays practical application face challenge about accuracy and consistency. Therefore, this article proposes a DT driven soft sensing method for key variables in rotary kiln. First, a thermodynamics and chemical reactions coupling model is built after analyzing the mechanism in kiln. Second, key parameters of DT were identified through analysis of limited and multisource data to ensure its consistency. Finally, in DT deployment stage, to realize the DT soft-sensing real-timely, a model reduction and DT distributed computing method was proposed to improve simulation timeliness. Practical application proved that soft sensing method based on DT can accurately and effectively obtain real-time monitoring results of key variables in the rotary kiln.

With the digital transformation of process manufacturing, identifying the system model from process data and then applying to predictive control has become the most dominant approach in process control. However, the controlled plant often operates under changing operating conditions. What is more, there are often unknown operating conditions such as first appearance operating conditions, which make traditional predictive control methods based on identified model difficult to adapt to changing operating conditions. Moreover, the control accuracy is low during operating condition switching. To solve these problems, this article proposes an error-triggered adaptive sparse identification for predictive control (ETASI4PC) method. Specifically, an initial model is established based on sparse identification. Then, a prediction error-triggered mechanism is proposed to monitor operating condition changes in real time. Next, the previously identified model is updated with the fewest modifications by identifying parameter change, structural change, and combination of changes in the dynamical equations, thus achieving precise control to multiple operating conditions. Considering the problem of low control accuracy during the operating condition switching, a novel elastic feedback correction strategy is proposed to significantly improve the control accuracy in the transition period and ensure accurate control under full operating conditions. To verify the superiority of the proposed method, a numerical simulation case and a continuous stirred tank reactor (CSTR) case are designed. Compared with some state-of-the-art methods, the proposed method can rapidly adapt to frequent changes in operating conditions, and it can achieve real-time control effects even for unknown operating conditions such as first appearance operating conditions.

Industrial system monitoring includes fault diagnosis and anomaly detection, which have received extensive attention, since they can recognize the fault types and detect unknown anomalies. However, a separate fault diagnosis method or anomaly detection method cannot identify unknown faults and distinguish between different fault types simultaneously; thus, it is difficult to meet the increasing demand for safety and reliability of industrial systems. Besides, the actual system often operates in varying working conditions and is disturbed by the noise, which results in the intraclass variance of the raw data and degrades the performance of industrial system monitoring. To solve these problems, a metric learning-based fault diagnosis and anomaly detection method is proposed. Fault diagnosis and anomaly detection are adaptively fused in the proposed end-to-end model, where anomaly detection can prevent the model from misjudging the unknown anomaly as the known type, while fault diagnosis can identify the specific type of system fault. In addition, a novel multicenter loss is introduced to restrain the intraclass variance. Compared with manual feature extraction that can only extract suboptimal features, it can learn discriminant features automatically for both fault diagnosis and anomaly detection tasks. Experiments on three-phase flow (TPF) facility and Case Western Reserve University (CWRU) bearing have demonstrated that the proposed method can avoid the interference of intraclass variances and learn features that are effective for identifying tasks. Moreover, it achieves the best performance in both fault diagnosis and anomaly detection.

Multimode process monitoring plays a significant role in ensuring the stable operation of industrial processes under changing conditions. Due to the continuous emergence of new modes, some adaptive model updating methods are proposed. However, the updated model may forget important features learned in previous modes, thus reducing the monitoring performance. To address this problem, this article proposes a global information-based lifelong dictionary learning (GI-LDL) method for multimode process monitoring. Specifically, this article adopts dictionary atoms as the fundamental units for representing process data and proposes a method for measuring the importance of dictionary atoms based on global information. Then, to ensure that the dictionary retains its representational ability for both new and historical mode data during the mode updating process, a surrogate quadratic loss considering the importance is further proposed to penalize changes of important atoms. Compared with dictionary constraints, finer-grained atomic constraints ensure that the dictionary preserves important features of previous modes while learning features of new modes. Finally, considering that the number of modes in multimode industrial processes is often unknown in advance, this article explicitly derives analytical solutions for dictionary updating, thus it is capable of accommodating ever-increasing modes in real industrial processes. To verify the effectiveness and advancement of the proposed method, extensive experiments are elaborately designed, and experimental results indicate that the proposed method has precise monitoring capabilities for both historical and new modes.

Modern industrial process control suffers from various difficulties, such as multivariable, multiconstrained, multiobjective, and strong nonlinearity. Model predictive control (MPC) is an effective solution and is widely used in industrial processes. However, one limitation of MPC is that sufficient data are required to build accurate predictive models. To this end, this article proposes a knowledge-informed neural network MPC solution. First, a Hammerstein system structure knowledge extraction method based on sparse representation is proposed, which is able to extract system structure knowledge from a small amount of system operation data. Then, a knowledge-informed neural network model is designed, which combines the system structure knowledge to construct a neural network with a special structure, thus overcoming the problem of insufficient data during the model training. Finally, the knowledge-informed neural network model is embedded in the MPC framework, which can reduce the computational cost of rolling optimization while ensuring prediction performance. A numerical simulation and a pH neutralization process experiment are conducted to verify the feasibility and effectiveness of the proposed method.

Modern industrial processes are essentially complex multivariable systems with high interaction among variables. Moreover, in modern process control, it is imperative not only to track a setting point for the controlled object but also the performance indicators for the entire operation process, where the optimal operational feedback control of multivariable industrial processes is derived. In this paper, we propose an operation output-feedback predictive control (OOFPC) method for multivariable industrial processes based on model order reduction and model predictive optimization. Firstly, causality networks are employed to extract potential correlations between variables to reduce the high-dimensional feature space. Subsequently, a setpoint compensator is developed within the framework of model predictive control (MPC), taking into account both performance indicators and operational constraints. Finally, closed-loop control with stable adjustment of the controlled object is achieved by utilizing proportional-integral-derivative (PID) controllers. Feedback correction and rolling optimization enable optimal setpoint tracking and stable closed-loop control. Experiments on a grinding process control case demonstrate the effectiveness of the proposed control strategy.

Graph representation learning (GRL) is a powerful tool for graph analysis, which has gained massive attention from both academia and industry due to its superior performance in various real-world applications. However, the majority of existing works for GRL are dedicated to node-based tasks and thus focus on producing node representations. Despite such methods can be used to derive edge representations by regarding edges as nodes, they suffer from sub-par result utility in practical edge-wise applications, such as financial fraud detection and review spam combating, due to neglecting the unique properties of edges and their inherent drawbacks. Moreover, to our knowledge, there is a paucity of research devoted to edge representation learning. These methods either require high computational costs in sampling random walks or yield severely compromised representation quality because of falling short of capturing high-order information between edges. To address these challenges, we present TER and AER, which generate high-quality edge representation vectors based on the graph structure surrounding edges and edge attributes, respectively. In particular, TER can accurately encode high-order proximities of edges into low-dimensional vectors in a practically efficient and theoretically sound way, while AER augments edge attributes through a carefully-designed feature aggregation scheme. Our extensive experimental study demonstrates that the combined edge representations of TER and AER can achieve significantly superior performance in terms of edge classification on 8 real-life datasets, while being up to one order of magnitude faster than 16 baselines on large graphs.

Smart contracts are programmable units that possess the ability to execute a wide range of computational tasks, operating on a decentralized ledger called a blockchain, frequently employed for the management of valuable digital assets. Unlike conventional programs, once smart contracts are deployed, they are immutable and cannot be altered. As the value associated with smart contracts increases, they become increasingly enticing targets for potential attackers. It is therefore crucial to conduct comprehensive testing of smart contracts prior to their deployment. Fuzzing is an important testing approach. Regrettably, existing coverage-based fuzzing tools treat all covered code in the same manner and cannot perform extensive tests on specific code fragments. In this research, we present LDfuzz, a targeted greybox fuzzer explicitly developed for Ethereum smart contracts. Its primary goal is to generate inputs that efficiently navigate towards potentially suspicious program locations. We propose suspicious branch marked, an innovative approach that assesses branch-level security implications. Utilizing the suggested metrics as a foundation, we calculate the distance of each fuzzing inputs to the marked branch and formulate a power scheduling algorithm based on simulated annealing, which progressively allocates additional energy to seeds in proximity to the suspicious branch while diminishing energy for seeds that are more distant. We assess the efficacy of LDfuzz by conducting a comparative analysis with a leading fuzzer for smart contracts, benchmarking their respective performance. The results from our experimentation findings demonstrate that LDfuzz exhibits superior efficiency compared to current advanced tools in identifying bugs within real-world contracts. Moreover, LDfuzz excels beyond current tools in achieving broader branch coverage.

Graph Neural Networks (GNNs) have shown superior performance for semi-supervised learning of numerous web applications, such as classification on web services and pages, analysis of online social networks, and recommendation in e-commerce. The state of the art derives representations for all nodes in graphs following the same diffusion (message passing) model without discriminating their uniqueness. However, (i) labeled nodes involved in model training usually account for a small portion of graphs in the semi-supervised setting, and (ii) different nodes locate at different graph local contexts and it inevitably degrades the representation qualities if treating them undistinguishedly in diffusion.

Abstract Channel and spatial attention mechanisms have proven to provide an evident performance boost of deep convolution neural networks. Most existing methods focus on one or run them parallel (series), neglecting the collaboration between the two attentions. In order to better establish the feature interaction between the two types of attentions, a plug‐and‐play attention module is proposed, which is termed as ‘CAT’—activating the Collaboration between spatial and channel Attentions based on learned Traits. Specifically, traits are represented as trainable coefficients (i.e. colla‐factors) to adaptively combine contributions of different attention modules to fit different image hierarchies and tasks better. Moreover, the global entropy pooling is proposed apart from global average pooling and global maximum pooling (GMP) operators, which is an effective component in suppressing noise signals by measuring the information disorder of feature maps. A three‐way pooling operation is introduced into attention modules and the adaptive mechanism is applied to fuse their outcomes. Extensive experiments on MS COCO, Pascal‐VOC, Cifar‐100, and ImageNet show that our CAT outperforms the existing state‐of‐the‐art attention mechanisms in object detection, instance segmentation, and image classification. The model and code will be released soon.

Smart manufacturing has become mainstream in the development of manufacturing industry, where Industrial Internet of Things plays a critical role. In this article, a systematic intelligent technique for procurement supply chain (PSC) optimization is proposed. In this technique, an integrated approach based on variational mode decomposition and long short-term memory network is used to predict the market price. Considering the factors, such as production plan and market fluctuation, a multiperiod dynamic purchasing model is built. A stacked autoencoder under bootstrap aggregation is then trained to evaluate suppliers automatically end-to-end based on various data. Finally, a multiobjective order allocation model is established considering the procurement costs and supplier scores, and solved by particle swarm optimization. The extensive experiments are performed using a realistic industrial application in a zinc smelter company. The experimental results demonstrate that the proposed technique greatly reduces labor costs, improves the efficiency of PSC, and reduces the procurement costs of the company.

Industrial cyber-physical systems (ICPSs) are deployed in many high-value facilities recently, and the monitoring of ICPS is more and more important. However, the prerequisite of ICPS monitoring is how to obtain an accurate network structure. In addition, the structure of ICPS may change over time and the observations data are limited and noisy. These situations make the ICPS network structure identification more difficult. In this article, we proposed the algorithm of temporal network identification from sparse data (ATNISD) to address these two issues simultaneously. First, we established the temporal network analysis model from the aspect of state equation and observation equation. Then, we analyze the characteristics of temporal networks in both time domain and space domain and propose a general framework of temporal networks structure identification, which is a combinatorial optimization problem. To improve the accuracy and alleviate the computational complexity, we decompose the combinatorial problem into small independent simple problems, which can be solved efficiently. The performance of the proposed algorithm is verified on synthetic evolutionary game dynamics on both homogeneous and heterogeneous temporal networks. The experimental results show that the proposed method can efficiently solve the problem of temporal networks structure identification from sparse data. Note to Practitioners—This article addresses the importance of network structure identification in industrial cyber-physical systems (ICPSs). The proposed method can effectively cope with the task of ICPS network structure identification in time-varying environments by exploiting the spatial and temporal features of networks in both time domain and space domain. The proposed algorithm can be implemented in typical slowing changing ICPS with or without observation noise, and it can decompose the combinatorial problem into small independent simple problems to improve the accuracy and release the computational complexity. Extensive simulation experiments demonstrate the accuracy and robustness of the proposed method for solving the structure identification task of temporal networks.

In real industrial processes, factors, such as the change in manufacturing strategy and production technology lead to the creation of multimode industrial processes and the continuous emergence of new modes. Although the industrial SCADA system has accumulated a large amount of historical data, which can be used for modeling and monitoring multimode processes to a certain extent, it is difficult for the model learned from historical data to adapt to emerging modes, resulting in the model mismatch. On the other hand, updating the model with data from new modes allows the model to continuously match the new modes, but it may cause the model to lose the ability to represent the historical modes, resulting in "catastrophic forgetting." To address these problems, this article proposed a jointly mode-matching and similarity-preserving dictionary learning (JMSDL) method, which updated the model by learning the data of new modes, so that the model can adaptively match the newly emerged modes. At the same time, a similarity metric was put forward to guarantee the representation ability of the proposed method for historical data. A numerical simulation experiment, the CSTH process experiment, and an industrial roasting process experiment indicated that the proposed JMSDL method can match new modes while maintaining its performance on the historical modes accurately. In addition, the proposed method significantly outperforms the state-of-the-art methods in terms of fault detection and false alarm rate.

Process monitoring, a typical application of industrial Internet of Things (IIOT), is crucial to ensure the reliable operation of the industrial system. In practice, due to the harsh environment and unreliable sensors and actuators, it is often difficult for IIoT to collect enough tagged and highly reliable data, which further degrades the process monitoring performance and makes the monitoring results not trustworthy. In order to reduce the negative impact of these unreliable factors, a self-weighted dictionary learning process monitoring method is proposed. In particular, a label propagation classifier is implemented from the labeled data to unlabeled data to obtain a credible label prediction. Subsequently, considering the interference of low-quality data and label information, we reweight the classification loss and label-consistency constraints to enhance the trustworthiness of feature extraction. Finally, a novel iterative optimization algorithm that combines the block coordinate descent method with the alternating direction multiplier method is developed to ensure the convergence speed of the learned classifier and dictionary. Extensive experiments indicate that the proposed method can guarantee the trustworthiness of the process monitoring results.

Health indicator prediction, such as remaining useful life prediction and product quality prediction, is an important aspect of industrial intelligence. It is essential to process the massive multichannel industrial time series collected from the Industrial Internet of Things for the industrial health indicator prediction. At present, there are still three issues that need to be considered for industrial health indicator prediction. First, it is difficult to directly connect the distant positions in the industrial time series to extract the temporal relations, which decreases the efficiency of extracting the potential long-distance temporal relations and training networks. Second, it should be fully considered that data from different channels have different contributions. Equally dealing with the contributions of each channel will weaken the representational ability of prediction networks. Third, the loss function deals with early predictions and delay predictions equally, which will lead to high risks caused by delay predictions. In this article, for these issues, a novel multichannel temporal attention-based network (MCTAN) is proposed for industrial health indicator prediction, which can weigh contributions of different channels through the channel attention while avoiding the loss of the temporal information and directly connect each time series position to the local fields of the sequence through the multi-head local attention mechanism to efficiently extract potential long-distance temporal relations. Then, a weighted mean square error loss function differently dealing with early predictions and delay predictions by setting dynamic weights is presented to reduce delay predictions. Next, to deal with the above-mentioned issues systematically, a framework combining data preprocessing and MCTAN collaboratively is introduced to predict industrial health indicators through multichannel time series. Finally, the experiments are carried out on the commercial modular aero-propulsion system simulation dataset to measure the performances, including the accuracy of industrial health indicator predictions and the inference speed.

Given a bipartite graph G consisting of inter-set weighted edges connecting the nodes in two disjoint sets U and V, bipartite network embedding (BNE) maps each node ui in U and vj in V to compact embedding vectors that capture the hidden topological features surrounding the nodes, to facilitate downstream tasks. Effective BNE should preserve not only the direct connections between nodes but also the multi-hop relationships formed alternately by the two types of nodes in G, which can incur prohibitive overheads, especially on massive bipartite graphs with millions of nodes and billions of edges. Existing solutions are hardly scalable to massive bipartite graphs, and often produce low-quality results. This paper proposes GEBE, a generic BNE framework achieving state-of-the-art performance on massive bipartite graphs, via four main algorithmic designs. First, we present two generic measures to capture the multi-hop similarity/proximity between homogeneous/heterogeneous nodes respectively, and the measures can be instantiated with three popular probability distributions, including Poisson, Geometric, and Uniform distributions. Second, GEBE formulates a novel and unified BNE objective to preserve the two measures of all possible node pairs. Third, GEBE includes several efficiency designs to get high-quality embeddings on massive graphs. Finally, we observe that GEBE achieves the best performance when instantiating MHS and MHP using a Poisson distribution, and thus, we further develop GEBEp based on Poisson-instantiated MHS and MHP, with non-trivial efficiency optimizations. Extensive experiments, comparing 15 competitors on 10 real datasets, demonstrate that our solutions, especially GEBEp, obtain superior result utility than all competitors for top-N recommendation and link prediction, while being up to orders of magnitude faster.

With the development of information and communication technologies, industrial cyber–physical systems (ICPSs) have accumulated a large amount of data, which enables us to convert data into industrial insight. However, since the industrial process of ICPS is always complicated and large scale, the raw data only contain a few operation condition information, which brings challenges to process monitoring and control. Thus, an efficient operation condition division method for ICPS is necessary. Although many operation condition division methods have been proposed, they were mainly relying on the static characteristics and ignored how the industrial process varies dynamically. Meanwhile, with the industrial process running, there may exist some new operation conditions that make the operation condition division task even more difficult. In order to grasp the static and dynamic features simultaneously of the industrial process and divide operation conditions accurately, we proposed an operation condition division method based on joint static and dynamic analysis with incremental learning. In detail, the slow feature analysis (SFA) and self-organizing map (SOM) network were proposed to extract the static and dynamic features jointly. Then, a division strategy was proposed to distinguish the operation condition changing points. For the new operating condition, we designed an incremental learning method based on the SOM network, which can update the operation condition model in real time. Extensive experiments, including a numerical simulation, two benchmark processes, and an industrial roasting process demonstrate that the proposed method can identify the operation conditions of the raw data in ICPS accurately and efficiently.

The development of industrial cyber-physical system (ICPS) provides a tight connection between the digital model and the industrial physical plant, which enables to use the data-driven methods to reflect the state of process running in the real world. However, due to the noisy and harsh industrial environment, the collected data are often corrupted to some extent. If the corrupted data are not detected in time, the data-driven model will inevitably degenerate and induce a poor process monitoring performance. In addition, the nonlinear characteristics between process variables due to the high complexities in physical plant bring challenges to the data-driven methods. In this article, a robust kernel dictionary learning method, which can overcome the negative influence of outliers and simultaneously extracts the nonlinear characteristics of industrial process, is proposed to address the above problems in ICPS. Our extensive experiments demonstrate that the proposed method has achieved significantly better and stable performance to deal with outlier detection and process monitoring in ICPS.Note to Practitioners—In order to mitigate the impacts due to process noise and outliers, a robust kernel dictionary learning method is proposed to improve the accuracy and stability of the process monitoring of industrial cyber-physical systems. This method considers the process noise, sparse outlier, as well as the nonlinear characteristic of industrial systems for improving the accuracy and stability of monitoring. Compared with many state-of-the-art methods, the proposed method can detect the outliers in the training dataset adaptively, which is more applicable to the real industrial system.

As one of the most effective technologies for network reconstruction, compressive sensing can recover signals from a small amount of observed data through sparse search or greedy algorithms in the assumption that the unknown signal is sufficiently sparse on a specific basis. However, there often occurs loss of precision even failure in the process of reconstruction without enough prior information. Therefore, the purpose of this article is to solve the problem of low reconstruction accuracy by mining implicit structural information in the network. Specifically, we propose a novel and efficient algorithm (MCM_TRA) for reconstructing the structure of the K -forked tree network. Based on evolutionary game dynamics, the modified clustering method (MCM) classifies all nodes into two sets, then a two-stage reconstruction algorithm (TRA) is illustrated to recover the node signals in different sets. The experimental results demonstrate that the MCM_TRA enhances the reconstruction accuracy prominently than previous algorithms. Moreover, extensive sensitivity analysis shows that the reconstruction effect can be promoted for a broad range of parameters, which further indicates the superiority of the proposed method.

The development of cloud manufacturing enables data-driven process monitoring methods to reflect the real industrial process states accurately and timely. However, traditional process monitoring methods cannot update learned models once they are deployed to edge devices, which leads to model mismatch when confronted time-varying data. In addition, limited resources on the edge prevent it from deploying complex models. Therefore, this article proposes a novel cloud-edge collaborative process monitoring method. First, historical data of industrial processes are collected to establish a dictionary learning model and train the dictionary and classifier in the cloud. Then, the model is simplified and deployed to the edge. The edge layer monitors the process states, including fault detection and working condition recognition, and determines whether a model mismatch has occurred based on an error-triggered strategy. Both numerical simulation and industrial roasting process results verify the superiority of the proposed method.

Hydraulic systems are a class of typical complex nonlinear systems, which have been widely used in manufacturing, metallurgy, energy, and other industries. Nowadays, the intelligent fault diagnosis problem of hydraulic systems has received increasing attention for it can increase operational safety and reliability, reduce maintenance cost, and improve productivity. However, because of the high nonlinear and strong fault concealment, the fault diagnosis of hydraulic systems is still a challenging task. Besides, the data samples collected from the hydraulic system are always in different sampling rates, and the coupling relationship between the components brings difficulties to accurate data acquisition. To solve the above issues, a deep learning model with multirate data samples is proposed in this article, which can extract features from the multirate sampling data automatically without expertise, thus it is more suitable in the industrial situation. Experiment results demonstrate that the proposed method achieves high diagnostic and fault pattern recognition accuracy even when the imbalance degree of sample data is as large as 1:100. Moreover, the proposed method can increase about 10% diagnosis accuracy when compared with some state-of-the-art methods.

This paper studies two variants of the best arm identification (BAI) problem under the streaming model, where we have a stream of n arms with reward distributions supported on [0,1] with unknown means. The arms in the stream are arriving one by one, and the algorithm cannot access an arm unless it is stored in a limited size memory. We first study the streaming \epslion-topk-arms identification problem, which asks for k arms whose reward means are lower than that of the k-th best arm by at most \epsilon with probability at least 1-\delta. For general \epsilon \in (0,1), the existing solution for this problem assumes k = 1 and achieves the optimal sample complexity O(\frac{n}{\epsilon^2} \log \frac{1}{\delta}) using O(\log^*(n)) memory and a single pass of the stream. We propose an algorithm that works for any k and achieves the optimal sample complexity O(\frac{n}{\epsilon^2} \log\frac{k}{\delta}) using a single-arm memory and a single pass of the stream. Second, we study the streaming BAI problem, where the objective is to identify the arm with the maximum reward mean with at least 1-\delta probability, using a single-arm memory and as few passes of the input stream as possible. We present a single-arm-memory algorithm that achieves a near instance-dependent optimal sample complexity within O(\log \Delta_2^{-1}) passes, where \Delta_2 is the gap between the mean of the best arm and that of the second best arm.

In batched multi-armed bandit problems, the learner can adaptively pull arms and adjust strategy in batches. In many real applications, not only the regret but also the batch complexity need to be optimized. Existing batched bandit algorithms usually assume that the time horizon T is known in advance. However, many applications involve an unpredictable stopping time. In this paper, we study the anytime batched multi-armed bandit problem. We propose an anytime algorithm that achieves the asymptotically optimal regret for exponential families of reward distributions with $O(\log \log T \ilog^{\alpha} (T))$ \footnote{Notation \ilog^{\alpha} (T) is the result of iteratively applying the logarithm function on T for \alpha times, e.g., \ilog^{3} (T)=\log\log\log T.} batches, where $\alpha\in O_{T}(1)$. Moreover, we prove that for any constant c>0, no algorithm can achieve the asymptotically optimal regret within c\log\log T batches.

Given a graph G where each node is associated with a set of attributes, and a parameter k specifying the number of output clusters, k-attributed graph clustering (k-AGC) groups nodes in G into k disjoint clusters, such that nodes within the same cluster share similar topological and attribute characteristics, while those in different clusters are dissimilar. This problem is challenging on massive graphs, e.g., with millions of nodes and billions of attribute values. For such graphs, existing solutions either incur prohibitively high costs, or produce clustering results with compromised quality.

Given a set V , the problem of unconstrained submodular maximization with modular costs (USM-MC) asks for a subset S ⊆ V that maximizes f ( S ) - c ( S ), where f is a non-negative, monotone, and submodular function that gauges the utility of S , and c is a non-negative and modular function that measures the cost of S. This problem finds applications in numerous practical scenarios, such as profit maximization in viral marketing on social media. This paper presents ROI-Greedy, a polynomial time algorithm for USM-MC that returns a solution S satisfying [EQUATION], where S * is the optimal solution to USM-MC. To our knowledge, ROI-Greedy is the first algorithm that provides such a strong approximation guarantee. In addition, we show that this worst-case guarantee is tight , in the sense that no polynomial time algorithm can ensure [EQUATION], for any ϵ > 0. Further, we devise a non-trivial extension of ROI-Greedy to solve the profit maximization problem, where the precise value of f ( S ) for any set S is unknown and can only be approximated via sampling. Extensive experiments on benchmark datasets demonstrate that ROI-Greedy significantly outperforms competing methods in terms of the tradeoff between efficiency and solution quality.

Data-driven process monitoring methods have attracted many attentions and gained wide applications. However, the real industrial process data are much more complex which is characterized by multimode, high dimensional, corrupted, and less labeled data. In order to eliminate these unfavourable factors simultaneously, a semisupervised robust projective and discriminative dictionary learning method is proposed. First, a semisupervised strategy is introduced to label unsupervised training data. Then, by utilizing low-rank and sparse features of raw data and outliers, a robust decomposition method is used to obtain clean data. After that, a simultaneously projective and discriminative model is proposed to extracting the feature of the low-rank clean data. Finally, the projection matrix and global dictionary, as well as the threshold are obtained through iterative dictionary learning. This hybrid framework provides a robust model for process monitoring and mode identification, and its efficiency is demonstrated by both synthetic examples and real industrial process cases.

Given a social network G, the profit maximization (PM) problem asks for a set of seed nodes to maximize the profit, i.e., revenue of influence spread less the cost of seed selection. The target profit maximization (TPM) problem, which generalizes the PM problem, aims to select a subset of seed nodes from a target user set T to maximize the profit. Existing algorithms for PM mostly consider the nonadaptive setting, where all seed nodes are selected in one batch without any knowledge on how they may influence other users. In this paper, we study TPM in adaptive setting, where the seed users are selected through multiple batches, such that the selection of a batch exploits the knowledge of actual influence in the previous batches. To acquire an overall understanding, we study the adaptive TPM problem under both the oracle model and the noise model, and propose ADG and AddATP algorithms to address them with strong theoretical guarantees, respectively. In addition, to better handle the sampling errors under the noise model, we propose the idea of hybrid error based on which we design a novel algorithm HATP that boosts the efficiency of AddATP significantly. We conduct extensive experiments on real social networks to evaluate the performance, and the experimental results strongly confirm the superiorities and effectiveness of our solutions.

Most industrial systems frequently switch their operation modes due to various factors, such as the changing of raw materials, static parameter setpoints, and market demands. To guarantee stable and reliable operation of complex industrial processes under different operation modes, the monitoring strategy has to adapt different operation modes. In addition, different operation modes usually have some common patterns. To address these needs, this article proposes a structure dictionary learning-based method for multimode process monitoring. In order to validate the proposed approach, extensive experiments were conducted on a numerical simulation case, a continuous stirred tank heater (CSTH) process, and an industrial aluminum electrolysis process, in comparison with several stateof-the-art methods. The results show that the proposed method performs better than other conventional methods. Compared with conventional methods, the proposed approach overcomes the assumption that each operation mode of industrial processes should be modeled separately. Therefore, it can effectively detect faulty states. It is worth to mention that the proposed method can not only detect the faulty of the data but also classify the modes of normal data to obtain the operation conditions so as to adopt an appropriate control strategy.

Hidden Markov models (HMMs) underpin the solution to many problems in computational neuroscience. However, it is still unclear how to implement inference of HMMs with a network of neurons in the brain. The existing methods suffer from the problem of being nonspiking and inaccurate. Here, we build a precise equivalence between the inference equation of HMMs with time-invariant hidden variables and the dynamics of spiking winner-take-all (WTA) neural networks. We show that the membrane potential of each spiking neuron in the WTA circuit encodes the logarithm of the posterior probability of the hidden variable in each state, and the firing rate of each neuron is proportional to the posterior probability of the HMMs. We prove that the time course of the neural firing rate can implement posterior inference of HMMs. Theoretical analysis and experimental results show that the proposed WTA circuit can get accurate inference results of HMMs.

We study the min-cost seed selection problem in online social networks for viral marketing, where the goal is to select a set of seed nodes with the minimum total cost such that the expected number of influenced nodes in the network exceeds a predefined threshold. We propose several algorithms that outperform the previous studies both on the theoretical approximation ratio and on the experimental performance. In the case where the nodes have heterogeneous costs, our algorithms are the first bi-criteria approximation algorithms with polynomial running time and provable approximation ratio. In the case where the users have uniform costs, our algorithms achieve logarithmic approximation ratio and provable time complexity which is smaller than that of the existing algorithms in orders of magnitude. We conduct extensive experiments using real social networks. The experimental results show that, our algorithms significantly outperform the existing algorithms both on the total cost and on the running time, and also scale well to billion-scale networks.

As a dual problem of influence maximization, the seed minimization problem asks for the minimum number of seed nodes to influence a required number η of users in a given social network G. Existing algorithms for seed minimization mostly consider the non-adaptive setting, where all seed nodes are selected in one batch without observing how they may influence other users. In this paper, we study seed minimization in the adaptive setting, where the seed nodes are selected in several batches, such that the choice of a batch may exploit information about the actual influence of the previous batches. We propose a novel algorithm, ASTI, which addresses the adaptive seed minimization problem in $O\Big(\fracη \cdot (m+n) \varepsilon^2 łn n \Big)$ expected time and offers an approximation guarantee of $\frac(łn η+1)^2 (1 - (1-1/b)^b) (1-1/e)(1-\varepsilon) $ in expectation, where η is the targeted number of influenced nodes, b is size of each seed node batch, and $\varepsilon \in (0, 1)$ is a user-specified parameter. To the best of our knowledge, ASTI is the first algorithm that provides such an approximation guarantee without incurring prohibitive computation overhead. With extensive experiments on a variety of datasets, we demonstrate the effectiveness and efficiency of ASTI over competing methods.

In modern process industries, many parameters or states can be acquired with sensors, and these parameters or states often have a close relationship with operation conditions. Unfortunately, the process often operates under different modes, and labels thereof are often unknown. In practice, labeling for sampled data is expensive and time-consuming, so identifying the operation conditions of the industrial process is difficult. In addition, sampled data from the industrial system are always contaminated by outliers or noise. Therefore, a robust process monitoring method for the multimode process is particularly important and challenging. In this paper, a robust dictionary learning method is proposed for processes with multiple unknown modes. Firstly, by taking the sparsity of outliers into account, a robust dictionary learning method is proposed to identify and remove the outliers and noise in the sampled training data. Secondly, an iterative minimization algorithm is designed for solving the dictionary learning optimization program. Thirdly, based on the sparsity of the sparse code, we partition the sparse code into different clusters via spectral clustering method, and then the dictionary is divided into some sub-dictionaries according to the cluster results of sparse code. Lastly, when a new sample is generated, we reconstruct it under different sub-dictionaries, and the smallest dictionary reconstruction error is calculated as a classifier for process monitoring and fault detection. To evaluate the validity and effectiveness of the proposed monitoring approach, we conduct extensive experiments on a numerical simulation, the continuous stirred tank heater (CSTH) process, and an industrial aluminum electrolysis process, in comparison with several state-of-the-art methods. The experimental results demonstrate that the proposed method is able to provide satisfying monitoring results, and it is also robust to outliers in the sampled training data. It is worth mentioning that the proposed method is an unsupervised learning method, therefore, it is more suitable for the process monitoring of real industrial systems.

Given a social network G , the influence maximization (IM) problem seeks a set S of k seed nodes in G to maximize the expected number of nodes activated via an influence cascade starting from S. Although a lot of algorithms have been proposed for IM, most of them only work under the non-adaptive setting, i.e., when all k seed nodes are selected before we observe how they influence other users. In this paper, we study the adaptive IM problem, where we select the k seed nodes in batches of equal size b , such that the choice of the i -th batch can be made after the influence results of the first i - 1 batches are observed. We propose the first practical algorithms for adaptive IM with an approximation guarantee of 1 − exp(ξ − 1) for b = 1 and 1 − exp(ξ − 1 + 1/ e ) for b > 1, where ξ is any number in (0, 1). Our approach is based on a novel AdaptGreedy framework instantiated by non-adaptive IM algorithms, and its performance can be substantially improved if the non-adaptive IM algorithm has a small expected approximation error. However, no current non-adaptive IM algorithms provide such a desired property. Therefore, we further propose a non-adaptive IM algorithm called EPIC, which not only has the same worst-case performance bounds with that of the state-of-the-art non-adaptive IM algorithms, but also has a reduced expected approximation error. We also provide a theoretical analysis to quantify the performance gain brought by instantiating AdaptGreedy using EPIC, compared with a naive approach using the existing IM algorithms. Finally, we use real social networks to evaluate the performance of our approach through extensive experiments, and the experimental experiments strongly corroborate the superiorities of our approach.

Influence maximization is a combinatorial optimization problem that finds important applications in viral marketing, feed recommendation, etc. Recent research has led to a number of scalable approximation algorithms for influence maximization, such as TIM + and IMM , and more recently, SSA and D-SSA . The goal of this paper is to conduct a rigorous theoretical and experimental analysis of SSA and D-SSA and compare them against the preceding algorithms. In doing so, we uncover inaccuracies in previously reported technical results on the accuracy and efficiency of SSA and D-SSA , which we set right. We also attempt to reproduce the original experiments on SSA and D-SSA , based on which we provide interesting empirical insights. Our evaluation confirms some results reported from the original experiments, but it also reveals anomalies in some other results and sheds light on the behavior of SSA and D-SSA in some important settings not considered previously. We also report on the performance of SSA-Fix , our modification to SSA in order to restore the approximation guarantee that was claimed for but not enjoyed by SSA . Overall, our study suggests that there exist opportunities for further scaling up influence maximization with approximation guarantees.

In recent years, Dropbox, Google, and Microsoft have been competing in the market of consumer cloud storage (CCS) services. While once the key comparative metric, storage capacity per user has outgrown the needs of most users. Today, third-party applications based on CCS's RESTful Web APIs are becoming a primary way for users to utilize their expanded storage resources. Unfortunately, there is very little visibility into the performance of these Web APIs, even though they are primary determinants of the end user experience on these storage applications. In this paper, we report results from a comprehensive measurement study of the Web APIs of five popular CCS providers. Our results reveal significant differences and limitations in API performance, which result in performance bottlenecks visible to the user through the storage application. We analyze the underlying system designs of the five providers' Web APIs, and present the performance implications of their different design choices. Our research provides practical guidance for service providers to optimize their API performance, for developers to improve the experience of third-party applications, and for users to pick appropriate services that best match their requirements.