Li Huang, Zhongxin Liu, Yifan Wu, Tao Yin, Dong li, Jichao Bi, Nankun Mu, Hongyu Zhang, Meng Yan. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Time-series forecasting in the energy sector is a labor-intensive process requiring expertise in multiple areas, such as data preprocessing, feature engineering, and neural network optimization. Although large language models offer automation potential, existing large-language-model-based multi-agent systems lack specialized structures for forecasting workflows, balancing computational cost and tool capability remains challenging. To address these issues, we propose an Energy-domain End-to-end Automated Forecasting Multi-Agent System, an on-premises end-to-end forecasting framework built upon a novel Department-Collaborative Multi-agent System Structure, specifically fine-tuned for energy forecasting applications. Within the Energy-domain End-to-end Automated Forecasting Multi-Agent System dominated by Department-Collaborative Multi-agent System Structure, two core and innovative modules are introduced: (i) Structured Behavioral Knowledge Distillation, which enables small-parameter large language models to operate feature engineering tools, facilitating lightweight and efficient on-premises deployment; and (ii) the Alignment-to-Selection Framework, which automates neural network selection by integrating architectural knowledge with historical performance data. Extensive experimental results demonstrate that the novel Energy-domain End-to-end Automated Forecasting Multi-Agent System significantly reduces manual effort, while maintaining high forecasting accuracy across diverse datasets.

A physics-data-driven economic model predictive control (EMPC) is proposed in this article for effective energy harvesting in wave energy converters (WECs). By combining artificial intelligence techniques, this article develops a new method that applies a data-driven model upon physical WEC models to address the challenges associated with the nonlinearity and uncertainty in WEC physical models. By collecting the error data between the actual system and the physical model, a deep Koopman operator is applied to transform the nonlinear and uncertain parts of the actual system into a linear model, which is then embedded into the physical model to establish a physical-data-driven model. By iteratively optimizing the physical data-driven model, EMPC generates the optimal control sequence for the WEC. Theoretical analysis is conducted to prove that the physical-data-driven EMPC algorithm ensures that the Lyapunov function converges to the neighborhood of the optimal steady state. Simulation results show that the proposed physical-data-driven model achieves faster convergence and higher accuracy during training compared to data-driven models. This improves the system’s control and optimization performance under EMPC, demonstrating the effectiveness of the proposed algorithm.

Efficiently capturing the complex spatiotemporal representations from large-scale traffic data with uneven data quality remains to be a challenging task. In considering of the dilemma, this work employs the advanced contrastive learning and proposes a novel Spatial-Temporal Synchronous Contextual Contrastive Learning (STS-CCL) model. First, we elaborate the basic and strong augmentation methods for spatiotemporal graph data. Second, we introduce a Spatial-Temporal Synchronous Contrastive Module (STS-CM) to simultaneously capture the decent spatial-temporal dependencies and realize graph-level contrasting. To further discriminate node individuals in negative filtering, a Semantic Contextual Contrastive method is designed based on semantic features and spatial heterogeneity, achieving node-level contrastive learning along with negative filtering. Finally, we present a hard mutual-view contrastive training scheme and extend the classic contrastive loss to an integrated objective function, yielding better performance. Extensive experiments and evaluations demonstrate that building a predictor upon STS-CCL contrastive learning model gains superior performance than existing traffic forecasting benchmarks. The proposed STS-CCL is highly suitable for large datasets with only a few labeled data and other spatiotemporal tasks with data scarcity issue.

Data-driven fault diagnosis for critical industrial processes has exhibited promising potential with massive operating data from the supervisory control and data acquisition system. However, automatically extracting the complicated interactions between measurements and subtly integrating them with temporal evolutions have not been fully considered. Besides, with the increasing complexity of industrial processes, accurately locating fault roots is of tremendous significance. In this article, we propose an unsupervised spatial-temporal aware graph encoder–decoder (STAGED) model for industrial fault diagnosis. First, the high-dimensional measurements are constructed as a weighted graph to depict the complicated interactions. Then, the graph convolutional network, long short-term memory network and attention mechanism are applied to learn a comprehensive representation for multiseries. To enforce the model to better capture the temporal evolution, the dual decoder that performs reconstruction and prediction tasks simultaneously is adopted with a well-designed comprehensive loss function. By learning the spatial-temporal evolutions of datasets, faults can be diagnosed and located at a fine-grained level based on reconstruction deviations. To verify the performance of STAGED, experiments on the Cranfield three-phase flow facility and secure water treatment datasets are implemented and the results indicate that it can provide insight into fault evolution and accurately diagnose faults.

Consensus clustering methods provide better performance by fusing multiple clustering solutions in terms of accuracy, robustness and stability. However, most current methods suffer from different challenges: i) the high-dimensional problem; ii) limitations of single clustering method; iii) the optimal number of clusters selecting for a certain validity measure; iv) redundant clustering candidate attributes. To overcome the above limitations, we propose a hybrid clustering solutions fusion method based on gated three-way decision (HCFG) for data analysis. By integrating multiple clustering solutions and executing information fusion, HCFG enjoys four properties: (1) multiple random subspace generation strategy is utilized to generate diverse low-dimensional subspaces effectively; (2) a fusion framework that considers characteristics of both the soft clustering and hard clustering methods is designed, in which potential boundary of feature attribute sets is explored; (3) the optimal number of clusters is set by utilizing multiple clustering validity indices; (4) clustering solutions is considered as attributes and a gated three-way decision method is proposed to adaptively conduct attribute reductions. Extensive comparative experiments on 24 real-world data sets demonstrates the effectiveness and superiority of HCFG. Moreover, nonparametric tests are conducted to compare HCFG with multiple consensus clustering methods.

Industrial Internet of Things (IIoT) is vulnerable to advanced persistent threat (APT). In this article, we study a scenario in which APT is launched to attack IIoT devices. Considering the APTs lateral movement, a node-level state evolution model is established to calculate the probability of every device in an IIoT system to be compromised by APT. Based on this, a Stackelberg game model is proposed for the APT attacker and defender, which can accurately describe the gaming process. An effective computational approach is developed to obtain the potential Stackelberg equilibrium strategy pair of the game. Extensive case studies and comparison studies are conducted to validate the effectiveness of the proposed method.

With the emergence of Industry 5.0, it is significant to enable efficient cooperation between humans and machines in the Industrial Internet of Things (IIoT). However, achieving real-time and reliable transmission of data flows deriving from time-sensitive applications in IIoT remains an open challenge. In this paper, we propose a three-layer software-defined IIoT (SDIIoT) architecture to enable multiple industrial services and flexible network configuration. In particular, when network services change frequently in SDIIoT, the delay of the control plane has a great influence on the end-to-end delay of data flows. To address this issue, we portray two different service curves of OpenFlow switches to adapt to dynamic network status based on Network Calculus (NC). To elevate resource efficiency and comply with friendly environments, we minimize the total worst-case network cost under strict resource constraints and transmission requirements by exploiting the joint flow routing and scheduling algorithm (JFRSA). Our numerical simulation results demonstrate the effectiveness and efficiency of our solution.

Accurate citywide traffic flow prediction is of great importance to intelligent transportation system. Existing methods typically assume the complete citywide traffic data can be obtained in real-time, which is impossible in applications. Furthermore, many recent works only consider one single kind of spatial correlation in traffic network when building graph representations. This work proposes an adversarial learning framework named Multi-Graph Convolutional Generative Adversarial Networks (MGC-GAN) to address the aforementioned challenges. To generate citywide traffic flow predictions using limited traffic data, we construct three kinds of graphs using easily accessed geographical and semantic information to model the complex spatial correlations in citywide transportation networks. Following that, a parallel GCN layer is designed to separately process multiple graphs. In addition, we design the Parallel Graph Convolution and Temporal Convolution Module (PGTCM) to effectively capture the heterogeneous spatial-temporal dependencies. Extensive experiments are carried out on two citywide traffic datasets, demonstrating that MGC-GAN outperforms several state-of-the-art baseline methods.

Backboned by smart meter networks, Advanced Metering Infrastructures (AMIs) play a critical role in smart grids. This paper studies a new False Data Injection Attack (FDIA) scenario targeting AMIs, in which the attacker injects and propagates computer worms (i.e., false data codes) to maliciously increase the readings of networked smart meters and create economic loss to the end customers. This paper establishes the false data code propagation and attack models in such a scenario; based on this, this paper proposes a differential game model for describing the attack and defense process for FDIA against AMIs. A computationally efficient algorithm is developed to solve the proposed differential model and obtain the potential Nash equilibrium (NE) strategy pair. Extensive numerical simulations are conducted to validate the effectiveness of the proposed method under different energy tariff structures.

Deploying Intrusion Detection Systems (IDSs) is an essential way to enhance the security of Multi-hop Clustered Wireless Sensor Networks (MCWSNs). The conventional IDS deployment architectural designs show limitations in ensuring the security of MCWSNs due to the limited monitoring range of the nodes. This paper proposes an efficient IDS deployment architecture for MCWSNs. The architecture is a hybrid design, in which the cluster heads and the sink collaboratively act as IDS agents to monitor the entire network and perform intrusion detection. Based on the new architecture, this paper proposes a resource allocation model to optimally allocate resources among the IDS agents so that the network’s security metric can be maximized. A comprehensive analytical framework is proposed to analyze the optimality of the model’s solution, and an efficient computational approach is developed to obtain the optimal resource allocation strategy. Extensive numerical simulation and comparison studies are conducted to validate the proposed method.

As an emerging and prospective paradigm, the industrial Internet of Things (IIoT) enable intelligent manufacturing through the interconnection and interaction of industrial production elements. The traditional approach that transmits data in a single physical network is undesirable because such a scheme cannot meet the network requirements of different industrial applications. To address this problem, in this article, we propose a network slicing orchestration system for remote adaptation and configuration in smart factories. We exploit software-defined networking and network functions virtualization to slice the physical network into multiple virtual networks. Different applications can use a dedicated network that meets its requirements with limited network resources with this scheme. To optimize network resource allocation and adapt to the dynamic network environments, we propose two heuristic algorithms with the assistance of artificial intelligence and the theoretical analysis of the network slicing system. We conduct numerical simulations to learn the performance of the proposed algorithms. Our experimental results show the effectiveness and efficiency of our proposed algorithms when multiple network services are concurrently running in the IIoT. Finally, we use a case study to verify the feasibility of the proposed network slicing orchestration system on a real smart manufacturing testbed.