Class incremental learning (CIL) requires an agent to learn distinct tasks consecutively with knowledge retention against forgetting. Problems impeding the practice of CIL methods are twofold: (1) prompt update on non-i.i.d batch streams without boundary, namely the harsher online task-free CIL (OTCIL) scenario; (2) CIL methods suffer from heavy forgetting on learning long task streams, as shown in Fig. 1(a). To achieve efficient decision-making, the ensemble deep random vector functional link network (edRVFL) with forward regularization (-F) is proposed to replace the canonical Ridge (-R), reducing more regrets during OTCIL. Considering continuous distribution drifting on long stream, we further propose edRVFL-kF to adjust the intervention intensity of forward knowledge and derive incremental updates. edRVFL-kF can effectively avoid replay, retraining, and catastrophic forgetting while achieving lower regret over -R. Moreover, to improve robustness on non-i.i.d stream and eliminate intractable tuning of -kF, we rebuild with online Bayesian learning and propose the plug-and-play edRVFL-kF-Bayes, enabling all hard ks in multiple sub-learners to self-adapt to ever-changing distribution and optimization in OTCIL. Experiments were conducted on image datasets, including multiple evaluations, ablation tests, estimated forward, and compatibility studies, which distinctly validate the efficacy of edRVFL-kF-Bayes.

Online learning of deep neural networks faces challenges such as delayed non-incremental updating, increasing consumption, retrospective retraining, and catastrophic forgetting. To alleviate these drawbacks and achieve progressive immediate decision-making, we propose a novel Incremental Online Learning (IOL) framework of Randomized Neural Networks (Randomized NN), facilitating continuous improvements and analytics to Randomized NN performance in online scenarios. Within the framework, we further formulate IOL with ridge regularization (-R) and IOL with forward regularization (-F), both avoiding retrospective retraining and catastrophic forgetting. Moreover, the incremental algorithms for -R/-F on non-stationary batch stream are derived, featuring recursive weight updates and variable learning rates. Compared to -R, we recommend -F which improves learning performance using future unlabeled observations while further reducing online regrets to offline global experts. Additionally, we conduct a detailed analysis and theoretically derive relative cumulative regret bounds of the Randomized NN learners for -R/-F under adversarial assumptions via a novel methodology and present several corollaries, from which we observed the superiority in online learning acceleration and declined regret bounds of employing -F in IOL. Finally, our proposed methods were rigorously examined across diverse tasks, from simulation, regression, and classification tasks, to long-term time-series forecasting (LTSF) and continual learning (CL) fields, which distinctly validated the efficacy of the IOL frameworks and the advantages of forward regularization.

Forward regularization (-F) with unsupervised knowledge was advocated to replace canonical Ridge regularization (-R) in online linear learners, as it achieved a lower relative regret boundary. However, we observe that -F cannot perform as expected in practice, even possibly losing to -R for online tasks. We identify two main causes for this: (1) inappropriate intervened regularization, and (2) non-i.i.d. nature and data distribution changes in online learning (OL), both of which result in unstable posterior distribution and optima offset of the learner. To improve these, we first introduce the adjustable forward regularization (- k F), a more general -F with controllable knowledge intervention. We also derive - k F’s incremental updates with variable learning rate, and study relative regret and boundary in OL. Inspired by the regret analysis, to curb unstable penalties, we further propose - k F-Bayes style with k synchronously self-adapted to revise the intractable tuning of - k F by considering parametric posterior distribution changes in non-i.i.d. online data streams. Additionally, we integrate the - k F and - k F-Bayes into a multi-layer ensemble deep random vector functional link (edRVFL) and present two practical algorithms for batch learning, avoiding past replay and catastrophic forgetting. In experiments, we conducted tests on numerical simulation, tabular, and image datasets, where - k F-Bayes surpassed traditional -R and -F, highlighting the efficacy of ready-to-work - k F-Bayes and the great potentials of edRVFL- k F-Bayes in OL and continual learning (CL) scenarios. • - k F provides a more flexible unsupervised knowledge intervention for online learners. • We derive - k F’s incremental updates and study relative regret in OL. • We propose - k F-Bayes to consider parametric posterior changes in non-i.i.d. streams. • We integrate the - k F and - k F-Bayes into edRVFL and present two algorithms for CL.

Cyber-physical systems (CPSs), which integrate sensing, computation, and control, are widely deployed in safety-critical applications such as industrial monitoring and predictive maintenance. Industrial fault diagnosis systems (IFDSs), as a representative CPS, enable intelligent health monitoring but are increasingly vulnerable to stealthy adversarial attacks—carefully crafted input perturbations that mislead models while remaining imperceptible. These attacks compromise system safety, highlighting the urgent need for learning-based security control in CPS. This article proposes multilayered noisy mixup (MLNM)-frequency shift (FrSh), a theoretically grounded adversarial defense framework for IFDS under real-world conditions involving nonstationary signals and class imbalance. The framework integrates two components: multilayered noisy mixup (MLNM) and frequency shift (FrSh). MLNM injects structured additive and multiplicative noise during training and is theoretically proven to introduce first- and second-order regularization terms in the loss function, enhancing generalization and gradient stability. FrSh performs spectrum-aware transformations that preserve conjugate symmetry and spectral integrity, with formal guarantees of aliasing-free, reversible frequency mapping, facilitating effective learning of high-frequency diagnostic features. Extensive experiments on benchmark datasets demonstrate that MLNM-FrSh significantly improves adversarial robustness and standard accuracy compared to state-of-the-art baselines. This work contributes a reliable and theoretically principled solution for attack-resilient fault diagnosis, advancing the broader goal of intelligent security control in CPS.

Dynamic matrix control (DMC) has been field-validated in many industrial practices, highlighting the critical importance of fine-tuning parameters for optimal performance. However, the tuning of well-performed parameters is challenging because the relationship between parameters and the performance of DMC is intricate to characterize for industrial systems with uncertainty. An adaptive tuning approach based on deep reinforcement learning (DRL) is proposed to optimize the performance of DMC for uncertain systems in this paper. The approach can online tune the horizons and weighting matrices of DMC in real time adaptive to the state and uncertainty of the systems. Compared with offline tuning approaches, the proposed approach does not need to tradeoff optimality for robustness. The proposed approach utilizes various state-of-the-art DRL algorithms, e.g., value-based and actor-critic-based, to develop online parameter tuning policies that can adapt to system uncertainty. A piecewise reward function is designed to improve the performance and stability of the agent. A novel predictor-switching criterion is developed to address the horizon inconsistency in the receding optimization process. The proposed approaches are validated by the moisture control task in industrial cigarette drying process. Note to Practitioners—This paper is motivated by the adaptive tuning problem of dynamic matrix control (DMC) in uncertain industrial systems. For other nonlinear industrial scenarios, practitioners should first design a nonlinear model predictive controller suitable for the controlled object. Then, they can refer to the proposed tuning algorithm to improve the controller performance. Specifically, regarding the setting of state and action sets, please refer to the technical details provided in this paper. The reward function can be flexibly set according to the needs of practitioners for the controller, e.g., improving the dynamic performance of the controller or saving controller energy consumption. The design processes of the tuning algorithms can refer toAlgorithm 1andAlgorithm 2. There are two reasons why the proposed algorithm can be easily transferred to nonlinear industrial scenarios. First, the proposed algorithm does not restrict the model predictive controller or controlled object type. Second, the requirement for horizons or weighting matrices tuning widely exists in the practical applications of various model predictive algorithms.

Industrial fault diagnosis systems(IFDS) play a crucial role in industrial systems but are susceptible to cyberattacks with data manipulation. Existing security methods are costly and ineffective for stealthy attacks. To address these challenges, we propose a security framework that synergizes covert attacks and low-overhead defenses in this article. The framework unifies a fault diagnostic feature-based adversarial attack (FDFA) strategy and a certified bound-based representative sampling (CBRS) one. With the FDFA, adversarial examples are created to deceive the monitor in an IFDS. The CBRS strategy filters the representative examples by computing the certified bounds of the clean examples, and then generates adversarial examples that approximate the decision bounds. FDFA-CBRS balances adversarial robustness and standard accuracy while reducing computational cost. Extensive experiments on public datasets demonstrate FDFA-CBRS's advantage in improving the robustness of IFDS confrontation, maintaining the standardized accuracy, and reducing the computational cost.

This paper proposes a strong prescribed-time multiple targets hunting control scheme for marine vehicles, including the cooperative hunting guidance strategy and the prescribed-time control algorithm. A real-time task allocation mechanism is introduced to dynamically assign hunting vehicles to multiple targets. By integrating both current and predicted target states, a dynamic attractive potential field is established to guide hunting vehicles toward favorable hunting positions. To ensure navigation safety, a repulsive potential field is designed in compliance with the International Regulations for Preventing Collisions at Sea (COLREGs). An obstacle speed-regulated mechanism is proposed to dynamically adjust the repulsive influence region. Moreover, a global prescribed-time control algorithm is proposed based on the hybrid mapping scheme to achieve strong convergence. A finite and continuous gain function prevents the typical singularity problem. Furthermore, a prescribed-time disturbance observer is constructed to compensate for model uncertainties and disturbances, improving system robustness in dynamic ocean environments. Through Lyapunov theory, all signals of the closed-loop system are guaranteed to be globally uniformly bounded at the prescribed time. The experiment results demonstrate the effectiveness of the proposed target hunting control scheme.

Dynamic matrix control (DMC) is widely used in intelligent manufacturing. Its performance is heavily affected by coupled parameters, e.g., horizons and weighting matrices. However, existing methods only focus on tuning of single type of parameter and cannot jointly tune them to improve the performance of DMC comprehensively. To bridge the gap, we propose a framework to tune DMC parameters based on double-loop-optimization jointly. The horizons of prediction and control are tuned in the outer loop, while the weighting matrices of error and control are tuned in the inner loop. An improved genetic algorithm fused with the particle swarm optimization is developed to minimize the cost function considering the constraints on horizons and weighting matrices. In the developed algorithm, population evolution and individual updating are deeply integrated to improve global optimality. The efficacy of the proposed framework is verified with real data of moisture control in cigarette drying process, which improves the quality of the moisture control by stabilizing the production equipment and reducing energy consumption.

The paper investigates the distributed cooperative control problem for multiple marine vehicles in the presence of sensor faults and limited communication networks. The composite learning adaptive fault-tolerant control algorithm is designed via the novel prediction error to tackle the perturbation incurred by possible sensor faults. To the best of the authors’ knowledge, the application of composite adaptive control to deal with sensor faults is the first attempt for marine vehicle systems. Besides, composite neural networks (NNs) are constructed to reconstruct model uncertainties. Different from the existing schemes, a concise event-triggered communication mechanism is proposed to optimize inter-vehicle communication. In particular, the time-varying parameter is introduced to dynamically adjust the event-triggered virtual control law according to the feedback of real-time actual tracking error, thereby enhancing control accuracy. Only the attitude information between the multiple vehicle members is aperiodically exchanged at sampling instants, saving communication resources. Based on the Lyapunov criterion, the semi-global uniformly ultimately bounded (SGUUB) stability of the closed-loop system can be guaranteed covering both trigger instants and continuous intervals. Two experiment results are illustrated to verify the effectiveness of the proposed scheme.

Deep reinforcement learning (DRL) has been widely applied in autonomous exploration and mapping tasks, but often struggles with the challenges of sampling efficiency, poor adaptability to unknown map sizes, and slow simulation speed. To speed up convergence, we combine curriculum learning (CL) with DRL, and first propose a Cumulative Curriculum Reinforcement Learning (CCRL) training framework to alleviate the issue of catastrophic forgetting faced by general CL. Besides, we present a novel state representation, which considers a local egocentric map and a global exploration map resized to the fixed dimension, so as to flexibly adapt to environments with various sizes and shapes. Additionally, for facilitating the fast training of DRL models, we develop a lightweight grid-based simulator, which can substantially accelerate simulation compared to popular robot simulation platforms such as Gazebo. Based on the customized simulator, comprehensive experiments have been conducted, and the results show that the CCRL framework not only mitigates the catastrophic forgetting problem, but also improves the sample efficiency and generalization of DRL models, compared to general CL as well as without a curriculum. Our code is available at https://github.com/BeamanLi/CCRL_Exploration.

The use of ensemble approaches in the single-objective evolutionary algorithms is ubiquitous, but ensembles of multiobjective evolutionary algorithms (MOEAs) have achieved relatively little attention. On the other hand, manually selecting a suitable MOEA to solve an actual multiobjective optimization problem (MOP) is time-consuming and challenging. Therefore, developing a multiobjective hyperheuristic to allocate computational resources for multiple MOEAs in an intelligent approach is beneficial. In this work, an autoselection strategy of MOEAs based on the performance indicator (MOEAS-PI) is introduced to alleviate the abovementioned problem. In the MOEAS-PI, the performance of each constituent MOEA in the pool is assessed according to a real-time and comprehensive performance indicator, which contains both the current and future performances. The MOEAS-PI is able to easily choose the best performing MOEA during the evolutionary process. Also, it can enhance the robustness of MOEAs and reduce the application risk. The effectiveness of the MOEAS-PI is carefully evaluated on 23 MOPs. Simulation results demonstrate that the MOEAS-PI is an effective and efficient method to integrate the advantages of each individual algorithm. Finally, the MOEAS-PI is utilized to solve a translation control problem of an immersed tunnel element under current flow. Experimental results reveal that the MOEAS-PI is a reliable and effective optimization approach to solve actual MOPs.Note to Practitioners—Multiobjective optimization problems (MOPs) have been commonly found in various fields. However, a single MOEA cannot guarantee its sufficient robustness and adaptability in solving MOPs. Therefore, this study aims to propose a multiobjective hyperheuristic algorithm to improve the robustness of MOEAs. The performance of the proposed algorithm is tested on benchmark test functions and an actual MOP. The results show that the proposed approach can select a suitable MOEA to solve a particular type of MOPs during the evolutionary process and provide a solution set for decision-makers to control the translation of an immersed tunnel element under different objectives/operator environments.

Good performance of a controller in Model Predictive Control (MPC) system keeps the whole industrial process running well. Because of the complexity of the process, data-driven performance assessment approach, instead of model approach, becomes a popular topic. However, performance assessment is inaccurate when operating condition changes, because the performance benchmark should be different. This paper proposes an overall index to classify different operating conditions of real-time dataset. This index is the sum of two similarity factors by adding a weight value. One is the Principal Component Analysis (PCA) similarity factor and another is Bhattacharyya distance similarity factor. This index, considering both characteristic and spatial distance of datasets, identifies the operating condition that the real-time data belongs to. The effectiveness of this index is demonstrated in the case of simulation.

This paper focuses on the stability analysis of T-S fuzzy control systems. The artificial T-S model method is utilized with piecewise linear interpolations into the normalized fuzzy membership functions. The stability conditions are derived to a series of LMIs. Using piecewise linear interpolation functions, we obtain finite LMIs and only solve them at the interpolation points. Furthermore, because of the dramatically improved approximation accuracy of piecewise linear interpolations, the method presented in this paper can provide a wider stable region for T-S fuzzy control systems, compared with the approach with staircase (zero-order) interpolations. A simulation example is adopted to illustrate the advantage of the proposed method.

An optimal point-wise control method for parabolic distributed parameter systems is proposed to solve the problem of determining both the locations of point-wise controllers and the control which should be exerted on each controller. For a given number of point-wise controllers, the optimal point-wise control form is given by solving a quadratic cost control problem, and the controller locations and the control are determined by minimizing the quadratic control cost performance index. The result indicates the method proposed is effective at solving the optimal point-wise control problem for parabolic distributed parameter systems.

In actual industrial processes, the measurement data always contain noise. Therefore, it will affect the accuracy of modeling. Compare to type-1 fuzzy sets, the membership functions in type-2 fuzzy sets include primary membership function and secondary membership function. It provides additional degrees of freedom that make it possible to model uncertainties brought by the noise. In this paper, a type-2 T-S fuzzy model is presented to minimize the effect of measurement noise. Furthermore, the influence of the initial conditions is considered in the algorithm. The primary membership function is gained through an improved nearest-neighborhood clustering algorithm, and the secondary membership function is determined through GMM based on the sufficient statistics. The orthogonal least-squared algorithm is used to identify the consequent of the fuzzy rules. Finally, the simulation results are compared with those obtained from a type-1 T-S fuzzy modeling results and the superiority of the proposed approach is highlighted.

A fuzzy satisfactory clustering algorithm is presented in this paper. It starts with two cluster centers and increases a new center if necessary. During the clustering process, the former clustering information is fully used so that the convergence rate can be speed up. A system data set can be quickly divided into several satisfactory fuzzy clusters by this algorithm. A Takagi-Sugeno type fuzzy model can then be identified. For three typical pH processes, satisfactory simulation results are obtained. The effective performance of the modified clustering algorithm is quantitatively evaluated.