IEEE/CAA Journal of Automatica Sinica

Federated Experiments: Generative Causal Inference Powered by LLM-based Agents Simulation and RAG-based Domain Docking

De-Yu Zhou, Xiao Xue, Qun Ma, Chao Guo, Li-Zhen Cui, Yong-Lin Tian, Jing Yang, Fei-Yue Wang

2025, 12(7): 1301-1304. doi: 10.1109/JAS.2024.124671

Abstract(278) HTML (7) PDF(70)

Abstract:

System Identification in the Network Era: A Survey of Data Issues and Innovative Approaches

Qing-Guo Wang, Liang Zhang

2025, 12(7): 1305-1319. doi: 10.1109/JAS.2024.125109

Abstract(61) HTML (5) PDF(16)

Abstract:
System identification is a data-driven modeling technique that originates from the control field. It constructs models from data to mimic the behavior of dynamic systems. However, in the network era, scenarios such as sensor malfunctions, packet loss, cyber-attacks, and big data affect the quality, integrity, and security of the data. These data issues pose significant challenges to traditional system identification methods. This paper presents a comprehensive survey of the emergent challenges and advances in system identification in the network era. It explores cutting-edge methodologies to address data issues such as data loss, outliers, noise and nonlinear system identification for complex systems. To tackle the data loss, the methods based on imputation and likelihood-based inference (e.g., expectation maximization) have been employed. For outliers and noise, methods like robust regression (e.g., least median of squares, least trimmed squares) and low-rank matrix decomposition show progress in maintaining data integrity. Nonlinear system identification has advanced through kernel-based methods and neural networks, which can model complex data patterns. Finally, this paper provides valuable insights into potential directions for future research.

Machine Learning-Based Prediction of Depressive Disorders via Various Data Modalities: A Survey

Qiong Li, Xiaotong Liu, Xuecai Hu, Md Atiqur Rahman Ahad, Min Ren, Li Yao, Yongzhen Huang

2025, 12(7): 1320-1349. doi: 10.1109/JAS.2025.125393

Abstract(59) HTML (5) PDF(10)

Abstract:
Depression, a pervasive mental health disorder, has substantial impacts on both individuals and society. The conventional approach to predicting depression necessitates substantial collaboration between health care professionals and patients, leaving room for the influence of subjective factors. Consequently, it is imperative to develop a more efficient and accessible prediction methodology for depression. In recent years, numerous investigations have delved into depression prediction techniques, employing diverse data modalities and yielding notable advancements. Given the rapid progression of this domain, the present article comprehensively reviews major breakthroughs in depression prediction, encompassing multiple data modalities such as electrophysiological signals, brain imaging, audiovisual data, and text. By integrating depression prediction methods from various data modalities, it offers a comparative assessment of their advantages and limitations, providing a well-rounded perspective on how different modalities can complement each other for more accurate and holistic depression prediction. The survey begins by examining commonly used datasets, evaluation metrics, and methodological frameworks. For each data modality, it systematically analyzes traditional machine learning methods alongside the increasingly prevalent deep learning approaches, providing a comparative assessment of detection frameworks, feature representations, context modeling, and training strategies. Finally, the survey culminates with the identification of prospective avenues that warrant further exploration. It provides researchers with valuable insights and practical guidance to advance the field of depression prediction.

Multi-UAV Cooperative Pursuit Strategy With Limited Visual Field in Urban Airspace: A Multi-Agent Reinforcement Learning Approach

Zhe Peng, Guohua Wu, Biao Luo, Ling Wang

2025, 12(7): 1350-1367. doi: 10.1109/JAS.2024.124965

Abstract(152) HTML (5) PDF(37)

Abstract:
The application of multiple unmanned aerial vehicles (UAVs) for the pursuit and capture of unauthorized UAVs has emerged as a novel approach to ensuring the safety of urban airspace. However, pursuit UAVs necessitate the utilization of their own sensors to proactively gather information from the unauthorized UAV. Considering the restricted sensing range of sensors, this paper proposes a multi-UAV with limited visual field pursuit-evasion (MUV-PE) problem. Each pursuer has a visual field characterized by limited perception distance and viewing angle, potentially obstructed by buildings. Only when the unauthorized UAV, i.e., the evader, enters the visual field of any pursuer can its position be acquired. The objective of the pursuers is to capture the evader as soon as possible without collision. To address this problem, we propose the normalizing flow actor with graph attention critic (NAGC) algorithm, a multi-agent reinforcement learning (MARL) approach. NAGC executes normalizing flows to augment the flexibility of policy network, enabling the agent to sample actions from more intricate distributions rather than common distributions. To enhance the capability of simultaneously comprehending spatial relationships among multiple UAVs and environmental obstacles, NAGC integrates the “obstacle-target” graph attention networks, significantly aiding pursuers in supporting search or pursuit activities. Extensive experiments conducted in a high-precision simulator validate the promising performance of the NAGC algorithm.

A Learning-Based Passive Resilient Controller for Cyber-Physical Systems: Countering Stealthy Deception Attacks and Complete Loss of Actuators Control Authority

Liang Xin, Zhi-Qiang Long

2025, 12(7): 1368-1380. doi: 10.1109/JAS.2024.124683

Abstract(281) HTML (4) PDF(51)

Abstract:
Cyber-physical systems (CPSs) are increasingly vulnerable to cyber-attacks due to their integral connection between cyberspace and the physical world, which is augmented by Internet connectivity. This vulnerability necessitates a heightened focus on developing resilient control mechanisms for CPSs. However, current observer-based active compensation resilient controllers exhibit poor performance against stealthy deception attacks (SDAs) due to the difficulty in accurately reconstructing system states because of the stealthy nature of these attacks. Moreover, some non-active compensation approaches are insufficient when there is a complete loss of actuator control authority. To address these issues, we introduce a novel learning-based passive resilient controller (LPRC). Our approach, unlike observer-based state reconstruction, shows enhanced effectiveness in countering SDAs. We developed a safety state set, represented by an ellipsoid, to ensure CPS stability under SDA conditions, maintaining system trajectories within this set. Additionally, by employing deep reinforcement learning (DRL), the LPRC acquires the capacity to adapt and diverse evolving attack strategies. To empirically substantiate our methodology, various attack methods were compared with current passive and active compensation resilient control methods to evaluate their performance.

Prescribed Performance Control of Nonlinear Systems With Unknown Sign-Switching Virtual Control Coefficients

Jin-Zi Yang, Jin-Xi Zhang, Tianyou Chai

2025, 12(7): 1381-1390. doi: 10.1109/JAS.2025.125135

Abstract(52) HTML (8) PDF(27)

Abstract:
The problem of high-performance tracking control for the lower-triangular systems with unknown sign-switching virtual control coefficients as well as unmatched disturbances is investigated in this paper. Instead of the online estimation algorithm, the sliding mode method and the Nussbaum gain technique, a group of orientation functions are employed to handle the unknown sign-switching virtual control coefficients. The control law is combined with the orientation functions and the barrier functions lumped in a recursive manner. It achieves output tracking with the preassigned rate, overshoot, and accuracy. In contrast with the existing solutions, it is effective for the nearly model-free case, with the requirement for information of neither the system nonlinearities nor their bounding functions of the plant, nor the bounds of the disturbances. In addition, our controller exhibits significant simplicity, without parameter identification, disturbance estimation, function approximation, derivative calculation, dynamic surfaces, or command filtering. Two simulation examples are conducted to substantiate the efficacy and advantages of our approach.

Full Perception Head: Bridging the Gap Between Local and Global Features

Jie Hua, Zhongyuan Wang, Xin Tian, Qin Zou, Jinsheng Xiao, Jiayi Ma

2025, 12(7): 1391-1406. doi: 10.1109/JAS.2025.125333

Abstract(99) HTML (3) PDF(13)

Abstract:
Object detection is a fundamental task in computer vision that involves identifying and localizing objects within an image. Local features extracted by convolutions, etc., capture fine-grained details such as edges and textures, while global features extracted by full connection layers, etc., represent the overall structure and long-range relationships within the image. These features are crucial for accurate object detection, yet most existing methods focus on aggregating local and global features, often overlooking the importance of medium-range dependencies. To address this gap, we propose a novel full perception module (FP-Module), a simple yet effective feature extraction module designed to simultaneously capture local details, medium-range dependencies, and long-range dependencies. Building on this, we construct a full perception head (FP-Head) by cascading multiple FP-Modules, enabling the prediction layer to leverage the most informative features. Experimental results in the MS COCO dataset demonstrate that our approach significantly enhances object recognition and localization, achieving 2.7−5.7 AP_val gains when integrated into standard object detectors. Notably, the FP-Module is a universal solution that can be seamlessly incorporated into existing detectors to boost performance. The code will be released at https://github.com/Idcogroup/FP-Head.

Secure Consensus Control on Multi-Agent Systems Based on Improved PBFT and Raft Blockchain Consensus Algorithms

Jing Zhu, Chengfang Lu, Juanjuan Li, Fei-Yue Wang

2025, 12(7): 1407-1417. doi: 10.1109/JAS.2025.125300

Abstract(118) HTML (6) PDF(15)

Abstract:
There has been significant recent research on secure control problems that arise from the open and complex real-world industrial environments. This paper focuses on addressing the issue of secure consensus control in multi-agent systems (MASs) under malicious attacks, utilizing the practical Byzantine fault tolerance (PBFT) and Raft consensus algorithm in blockchain. Unlike existing secure consensus control algorithms that have strict requirements for topology and high communication costs, our approach introduces a node grouping methodology based on system topology. Additionally, we utilize the PBFT consensus algorithm for intergroup leader identity verification, effectively reducing the communication complexity of PBFT in large-scale networks. Furthermore, we enhance the Raft algorithm through cryptographic validation during followers’ log replication, which enhances the security of the system. Our proposed consensus process not only identifies the identities of malicious agents but also ensures consensus among normal agents. Through extensive simulations, we demonstrate robust convergence, particularly in scenarios with the relaxed topological requirements. Comparative experiments also validate the algorithm’s lower consensus latency and improved efficiency compared to direct PBFT utilization for identity verification and classical secure consensus control method mean subsequence reduced (MSR) algorithm.

Nonlinear Integral-Ameliorated Model for Dynamic Convex Optimization With Perturbance Considered

Kangze Zheng, Yunong Zhang

2025, 12(7): 1418-1433. doi: 10.1109/JAS.2024.124788

Abstract(124) HTML (4) PDF(17)

Abstract:
This work presents a nonlinear integral-ameliorated model for handling dynamic optimization problems with affine constraints. They pose a challenge as their optimal solutions evolve with time. Traditional iteration-based methods that exactly solve the problem at each time instant, fail to precisely and real-time track the solution due to computational and communication bottlenecks. Our model, through rigorous theoretical analyses, is able to reduce the optimality gap (i.e., the difference between the model state and optimal solution) to zero in a finite time, and thus, track the solution online. Besides, perturbance is taken into account. We prove that under certain conditions, our model can totally tolerate an important kind of noise that we call “error-related noise”. In numerical experiments, compared with six existing methods, our model exhibits superior robustness when contaminated by the error-related noise. The key techniques in the model design involve employing the zeroing neural network to leverage time-derivative information, and introducing an integral term as well as the class $ {{{\mathrm{C}}}^0_\text{L}} $ functions to enhance convergence and noise resistance. Finally, we establish a model-free control framework for a surgical manipulator with the remote-center-of-motion constraint and compare the performances of the framework based on different models in simulations. The results indicate that our model achieves the best performance among various models employed within the framework.

Online Estimation of DC-link Capacitor Parameters of Three-Level NPC Converters Using Inherent Signals Analysis

Ricardo Lucio de Araujo Ribeiro, Reuben Palmer Rezende de Sousa, Alexandre Cunha Oliveira, Antonio Marcus Nogueira Lima, Qing-Long Han

2025, 12(7): 1434-1444. doi: 10.1109/JAS.2025.125159

Abstract(158) HTML (6) PDF(22)

Abstract:
This paper presents a method for estimating the parameters of DC-link capacitors in three-level NPC voltage source inverters (3L-NPC-VSI) used in grid-tied systems. The technique uses the signals generated by the intermodulation caused by the PWM strategy and converter topology interaction to estimate the capacitor parameters of the converter DC-link. It utilizes an observer-based structure consisting of a recursive noninteger sliding discrete Fourier transform (rnSDFT) and an RLS filter improved with a forgetting factor (oSDFT-RLS) to accurately estimate the capacitance and equivalent series resistance (ESR). Importantly, this method does not require additional sensors beyond those already installed in off-the-shelf 3L-NPC-VSI systems, ensuring its noninvasiveness. Furthermore, the oSDFT-RLS estimates capacitor parameters in the time-frequency domain, enabling the tracking of capacitor degradation and predicting potential faults. Experimental results from the laboratory setup demonstrate the effectiveness of the proposed condition monitoring method.

An Intelligent Optimization Strategy for Blast Furnace Charging Operation Considering Three-Dimensional Burden Surface Shape

Jicheng Zhu, Zhaohui Jiang, Dong Pan, Haoyang Yu, Chuan Xu, Ke Zhou, Weihua Gui

2025, 12(7): 1445-1463. doi: 10.1109/JAS.2025.125192

Abstract(187) HTML (5) PDF(32)

Abstract:
Today, a well-devised charging operation scheme is urgently needed by on-site workmen and is critical for building an intelligent blast furnace (BF). Previous research on charging operations always focused on the two-dimensional shape of the burden surface (i.e., a single radial profile) while neglecting the unique feature of global dissymmetry, severely restricting the development of precise charging. For this reason, this study proposes an innovative optimization strategy for the charging operation under the three-dimensional burden surface, which is the first attempt in this field. First, a practicable region partitioning scheme is introduced, and the partitioning results are then integrated with the charging mechanism to construct a three-dimensional burden surface prediction model. Next, the intrinsic relationship between the operational parameters and charging volume is revealed based on the law of mass conservation, which forms the basis for defining a novel operational parameter with variable-speed utility, referred to as the neotype charging matrix (NCM). To find the best NCM, a customized NCM optimization strategy, involving a dual constraint handling technique in conjunction with a two-stage hybrid variable differential evolution algorithm, is further developed. The industrial experiment results manifest that the partitioning scheme significantly enhances the accuracy of burden surface description. Moreover, the NCM optimization strategy offers greater flexibility and higher accuracy than current mainstream optimization strategies for the charging matrix (CM).

Hybrid Event-Triggered Control With Stability Analysis

Ding Wang, Lingzhi Hu, Junfei Qiao

2025, 12(7): 1464-1474. doi: 10.1109/JAS.2024.125067

Abstract(54) HTML (2) PDF(26)

Abstract:
In this paper, a novel hybrid event-triggered control (ETC) method is developed based on the online action-critic technique, which aims at tackling the optimal regulation problem of discrete-time nonlinear systems. In order to ensure the normal execution of the online learning algorithm, a stability criterion condition is created to obtain the initial admissible control policy by using an offline iterative method under the time-triggered control framework. Subsequently, a general triggering condition is designed based on the uniform ultimate boundedness of the controlled system. In order to determine a constant interval which can ensure the system stability, another triggering condition is introduced and the asymptotic stability of the closed-loop system satisfying this condition is analyzed from the perspective of the input-to-state stability. The designed online hybrid ETC method not only further improves control efficiency, but also avoids the continuous judgment of the corresponding triggering condition. In addition, the event-based control law can approach the optimal control input within a finite approximation error. Finally, two experimental examples with physical background are conducted to indicate the present results.

A Robust Large-Scale Multiagent Deep Reinforcement Learning Method for Coordinated Automatic Generation Control of Integrated Energy Systems in a Performance-Based Frequency Regulation Market

Jiawen Li, Tao Zhou

2025, 12(7): 1475-1488. doi: 10.1109/JAS.2024.124482

Abstract(66) HTML (5) PDF(20)

Abstract:
To enhance the frequency stability and lower the regulation mileage payment of a multiarea integrated energy system (IES) that supports the power Internet of Things (IoT), this paper proposes a data-driven cooperative method for automatic generation control (AGC). The method consists of adaptive fractional-order proportional-integral (FOPI) controllers and a novel efficient integration exploration multiagent twin delayed deep deterministic policy gradient (EIE-MATD3) algorithm. The FOPI controllers are designed for each area based on the performance-based frequency regulation market mechanism. The EIE-MATD3 algorithm is used to tune the coefficients of the FOPI controllers in real time using centralized training and decentralized execution. The algorithm incorporates imitation learning and efficient integration exploration to obtain a more robust coordinated control strategy. An experiment on the four-area China Southern Grid (CSG) real-time digital system shows that the proposed method can improve the control performance and reduce the regulation mileage payment of each area in the IES.

A Transactional-Behavior-Based Hierarchical Gated Network for Credit Card Fraud Detection

Yu Xie, MengChu Zhou, Guanjun Liu, Lifei Wei, Honghao Zhu, Pasquale De Meo

2025, 12(7): 1489-1503. doi: 10.1109/JAS.2025.125243

Abstract(182) HTML (7) PDF(26)

Abstract:
The task of detecting fraud in credit card transactions is crucial to ensure the security and stability of a financial system, as well as to enforce customer confidence in digital payment systems. Historically, credit card companies have used rule-based approaches to detect fraudulent transactions, but these have proven inadequate due to the complexity of fraud strategies and have been replaced by much more powerful solutions based on machine learning or deep learning algorithms. Despite significant progress, the current approaches to fraud detection suffer from a number of limitations: for example, it is unclear whether some transaction features are more effective than others in discriminating fraudulent transactions, and they often neglect possible correlations among transactions, even though they could reveal illicit behaviour. In this paper, we propose a novel credit card fraud detection (CCFD) method based on a transaction behaviour-based hierarchical gated network. First, we introduce a feature-oriented extraction module capable of identifying key features from original transactions, and such analysis is effective in revealing the behavioural characteristics of fraudsters. Second, we design a transaction-oriented extraction module capable of capturing the correlation between users’ historical and current transactional behaviour. Such information is crucial for revealing users’ sequential behaviour patterns. Our approach, called transactional-behaviour-based hierarchical gated network model (TbHGN), extracts two types of new transactional features, which are then combined in a feature interaction module to learn the final transactional representations used for CCFD. We have conducted extensive experiments on a real-world credit card transaction dataset with an increase in average F1 between 1.42% and 6.53% and an improvement in average AUC between 0.63% and 2.78% over the state of the art.