This paper investigates task allocation and path planning in multi-agent pursuit-evasion (MPE) games under Dubins vehicle kinematics. A hierarchical framework is proposed: first, an iterative task allocation algorithm integrates time-optimal Dubins paths with bipartite graph matching to decompose the MPE game into parallel subgames. Subsequently, each subgame is formulated as a non-zero-sum differential game where pursuers exhibit altruistic behavior by incorporating teammates’ performance into their performance indices, dynamically optimizing team capture efficiency. An online adaptive dynamic programming (ADP) algorithm is developed to approximate Nash equilibrium policies in a distributed manner. Finally, simulations demonstrate that the proposed framework achieves complete evader capture, with altruistic behavior reducing capture time compared to self-interested strategies.

In this paper, the multi-agent formation control methods of differential graphical game against deception attacks are investigated. Firstly, based on the differential graphical game theory, we propose the sufficient conditions of formation establishment under potential attacks, and develop a passive attack-tolerant formation control scheme. Secondly, taking the attack as a player in games, we explore the formation strategy of active attack-tolerance in a zero-sum differential graphical game framework. Thirdly, we compare and analyze the two strategies in terms of formation error convergence speed and robustness to attacks theoretically. The proposed theoretical results are verified by a numerical simulation.

Resilience describes a networked system's ability to maintain fundamental functionality despite faults or errors. Existing resilience assessments focus on the static removal of nodes or edges from the topology, which limits predicting which node perturbation, at what moment, will critically threaten functionality. This paper presents a spectrum-entropy framework that quantifies functionality through the diversity of dynamical modes. The entropy degradation following a single-node perturbation unfolds across three temporal regimes: a local scale with degree-dependent impact, a critical scale with size-independent vulnerability governed by the Fiedler eigenvalue and dynamical parameters, and an asymptotic scale dependent on connectivity. Moreover, the most vulnerable node is determined by the largest Fiedler eigenvector component rather than node degree. At the critical time, infinitesimal perturbations at this node are amplified by the modal structure, driving peak entropy loss as the Fiedler mode alone dominates the dynamics. Both the critical time and the vulnerable node are computable offline, enabling preemptive intervention. Numerical simulations confirm the framework's effectiveness.

With increasing renewable integration into high-voltage direct current (HVDC) systems, transient overvoltages caused by commutation failures pose considerable challenges to the stable operation of wind farms. Conventional fixed-parameter excitation control in distributed synchronous condensers (DSCs) lacks adaptability and fails to fully utilize their dynamic reactive power capability. This study proposes a phase-staged excitation strategy based on the rate of change of voltage error, enabling real-time tuning of control parameters. A dual-channel control structure enhances adaptability to varying disturbance scenarios. Hardware-in-the-loop (HIL) simulation results show that the proposed method reduces peak overvoltage by 0.024 pu. and remarkably improves the voltage stability margin.

This paper develops a unified algebraic structural framework for analyzing global and local observ ability of network systems on non-uniform, directed, and weighted hypergraphs. By exploiting polynomial rings and Hilbert's basis theorem, a tensor representation encodes group-wise dynamics and measurements, yielding finitely generated Lie-derivative ideals. Necessary and sufficient global criteria are expressed through ordered hyperedge contractions, which can be viewed as a directed higher order propagation chain from states to outputs. Anincre mental output-design algorithm explains how the criteria may guide output selection without claiming complete ness. Structural observability is then studied via hyper edge reachability and dynamic-output hypergraph automor phisms. Finally, local observability conditions are developed whose rank tests explicitly reflect the higher-order hypergraph structure. The effectiveness of the proposed criteria is demonstrated through a competitive population model with third-order interactions.

The continuous increase in the share of high-penetration renewable energy has heightened the complexity of power system dynamics. At the same time, multiple protection measures designed to ensure the safe operation of doubly-fed induction generators may aggravate the risk of short-term voltage instability under severe grid faults. To address this issue, this paper proposes a preventive control strategy suitable for new-type power systems, aiming to effectively mitigate such potential risks. First, the causes of repeated low-voltage ride-through events in doubly-fed induction generators and their impact on system stability are thoroughly analyzed. Next, a preventive control model is formulated with the objective of minimizing regulation costs while incorporating multiple constraints including power flow, static voltage stability, and short-term voltage stability. To solve this model efficiently, a collaborative solution framework integrating a physics-guided neural network and a long short-term memory network is developed. By leveraging the loss-minimization property of the physics-guided neural network, the objective function and constraints are constructed into corresponding loss terms, which enables the generation of candidate control strategies that respect physical laws. The long short-term memory network is then employed to predict the system voltage stability margin under each candidate strategy. The gradient of the stability margin with respect to each control variable is calculated and converted into gradient signals that are fed back into the loss function, thereby correcting the decision output of the neural network according to the intrinsic physical evolution of the system and achieving fast and accurate computation of the optimal preventive control strategy. Simulation results demonstrate that the proposed method improves the average short-term voltage stability index of key buses by 73.95% in the 10-machine 39-bus system and by 59.28% in a practical case system, confirming its effectiveness and superiority in enhancing the short-term voltage stability of power systems.

This paper considers for the first time pursuit-evasion (PE) differential games with irrational perceptions of both pursuer and evader on probabilistic characteristics of environmental uncertainty. Firstly, the irrational perceptions of risk aversion and probability sensitivity are modeled and incorporated within a Bayesian PE differential game framework by using Cumulative Prospect Theory (CPT) approach; Secondly, several sufficient conditions of capturability are established in terms of system dynamics and irrational parameters; Finally, the existence of CPT-Nash equilibria is rigorously analyzed by invoking Brouwer's fixed-point theorem. The new results reveal that irrational behaviors benefit the pursuer in some cases and the evader in others. Certain captures that are unachievable under rational behaviors can be achieved under irrational ones. By bridging irrational behavioral theory with game-theoretic control, this framework establishes a rigorous theoretical foundation for practical control engineering within complex human-machine systems.

With the growing integration of renewable energy sources (RESs) and smart interconnected devices, conventional distribution networks have turned to active distribution networks (ADNs) with complex system model and power flow dynamics. The rapid fluctuation of RES power may easily result in frequent voltage violation issues. Taking the flexible RES reactive power as control variables, this paper proposes a two-layer control scheme with Koopman wide neural network (WNN) based model predictive control (MPC) method for optimal voltage regulation and network loss reduction. Based on Koopman operator theory, a data-driven WNN method is presented to fit a high-dimensional linear model of power flow. With the model, voltage and network loss sensitivities are computed analytically, and utilized for ADN partition and control model formulation. In the lower level, a dual-mode adaptive switching MPC strategy is put forward for optimal voltage control and network loss optimization in each individual partition to decide the RES reactive power. The upper level is to calculate the adjustment coefficients of the RES reactive power given in the low level by taking the coupling effects of different partitions into account, and then the final reactive power dispatches of RESs are obtained to realize optimal control of voltage and network loss. Simulation results on two ADNs demonstrate that the proposed strategy can reliably maintain the voltage at each node within the secure range, reduce network power losses, and enhance the overall system security and economic efficiency.

Multi-sensor measurement systems in oil wells face two key challenges in fault detection: existing methods fail to fully exploit the spatial topological information among sensors, and detection decisions lack confidence quantification, limiting the effective utilization of measurement information. To address these limitations, this paper proposes a dynamic spatio-temporal reasoning and confidence-driven gating model. A node integrity maintenance strategy is first designed to balance measurement data quality with topological completeness. A data preprocessing mechanism extracts “state” and “trend” feature sequences from raw signals, achieving dimensionality reduction and enhanced multi-source measurement data. On this basis, physical graph topology based on sensor layout is first introduced into this problem, and a centrality-guided graph convolution unit is proposed to model spatial relationships among sensors. In the temporal dimension, a dual-stream attention mechanism enhances system-level temporal measurement features through iterative refinement. Furthermore, a confidence-driven dynamic gating mechanism achieves efficient inference while ensuring detection confidence. The model is systematically validated on the 3W dataset, a real-world multi-sensor production dataset from Brazilian deepwater oil wells. Two representative faults with distinct evolution rates are selected — a rapid abrupt fault and a slow progressive fault — to evaluate generalization across different fault evolution patterns. Results demonstrate superior detection performance under both extreme modes. For the slow progressive fault, the model achieves an F1 score of 0.97 with a False Negative Rate of 0.02, while reducing inference cost by 20–25%, highlighting its practical value as a multi-sensor measurement-based intelligent detection scheme. Source code and datasets are available at https://github.com/CrazyFPGA/DSR-CDGNet.

A robust fault-tolerant trajectory tracking control strategy is proposed for underactuated hovercraft subject to actuator degradation, model uncertainties, and time-varying environmental disturbances. To overcome the restrictive requirement of conventional prescribed performance control (PPC) that the initial tracking error must remain within predefined bounds, a globally feasible prescribed performance design is developed by introducing auxiliary functions with adaptively regulated performance envelopes, thereby improving the applicability of PPC while preserving satisfactory transient and steady-state tracking performance. To address actuator faults such as thrust and yaw moment loss, an adaptive fault observer is designed to estimate actuator effectiveness degradation online and provide real-time compensation for the controller. In addition, a nonlinear disturbance observer is employed to reconstruct lumped uncertainties caused by modeling inaccuracies and external disturbances, so as to enhance the robustness of the closed-loop system. Based on these designs, a unified fault-tolerant prescribed performance tracking framework is established for underactuated hovercraft. Lyapunov-based analysis proves that all closed-loop signals remain bounded and that the system states are uniformly ultimately bounded. Numerical simulations demonstrate that the proposed method achieves improved tracking accuracy, faster convergence, and stronger robustness under large initial errors, actuator faults, and external disturbances.

This paper addresses the fault-tolerant control problem for multi-agent systems with flexible manipulators governed by coupled ordinary differential equation-partial differential equation (ODE-PDE) dynamics. A distributed control framework is proposed to handle actuator faults, flow-induced disturbances and obstacle avoidance constraints through three integrated mechanisms. First, each agent employs a modelbased baseline controller that compensates for unknown faults while suppressing elastic vibrations. Second, a neural disturbance perception encoder is designed to extract lowdimensional latent features from dynamics residuals, providing all agents with a consistent representation of the flow environment. Third, residual reinforcement learning policies trained with disturbance-aware rewards augment the baseline control through Riemannian motion policy (RMP) flow composition, ensuring asymptotic convergence to consensus under disturbance-free conditions and bounded tracking errors under persistent disturbances. Numerical simulations demonstrate the effectiveness of the proposed methods.

In this paper, the performance evaluation of feedback control systems under strictly stealthy attacks is studied. First, the existence condition and uniform form of strictly stealthy attacks against observe based feedback control systems are presented. Then, the estimation performance of feedback control systems under strictly stealthy attacks is evaluated through analyzing the effects of attacks on state estimation errors. The loop performance is selected as an index to evaluate the influence of stealthy attacks on control performance. In the attack-free case, the loop performance is decomposed as feedback robustness performance driven by residual and feedforward tracking performance driven by the normal reference signal. The strictly stealthy attacks only affect the feedforward tracking performance, via tampering the reference signal. Finally, simulations on an flight vehicle demonstrate the effectiveness of the proposed methods. (c) 2025 Elsevier Ltd. All rights are reserved, including those for text and data mining, AI training, and similar technologies.

This paper considers Nash equilibrium verification problem of inverse linear-quadratic (LQ) differential games, which answers the question whether a set of controllers constitutes a Nash equilibrium of a LQ differential game. This establishes a solid foundation for inverse differential games where performance indices can be identified through controllers. Distributed verification criteria are established via geometric characterization of constrained algebraic Riccati equation solutions, and can be easily checked from system dynamics and given feedback matrices. Moreover, an analytical reconstruction method is proposed for the corresponding state weighting matrices in the LQ performance indices.

In multi-parallel grid-following voltage source converters (MGFL-VSCs), the nonlinear dynamics of the control loops and their coupling interactions substantially affect the transient synchronization stability (TSS) of MGFL-VSCs. This study focuses on the effects of nonlinear coupling interactions on TSS of the MGFL-VSC systems. First, on the basis of differential–algebraic equation–stability region boundary (SRB) theory, the study analyzes the effects of nonlinear dynamics and coupling interactions and plots the SRB of MGFL-VSCs. Results show that the coupling interactions limit the maximum phase angle of the phase-locked loop. The SRB of MGFL-VSCs consists of a stable manifold of unstable equilibrium points, a stable manifold of semi-singular points, and certain singular surfaces. The composition of SRB is closely related to the coupling interactions of nonlinear control. Second, on the basis of the correlation between SRB and transient instability modes, the transient instability modes of MGFL-VSCs are divided into traditional and coupling interaction instability modes, and the evolution of instability modes is systematically studied in consideration of nonlinear coupling interactions. In addition, an online assessment method that is based on the Port–Hamiltonian energy function is proposed. This method can quantitatively estimate the effect of coupling interactions on TSS by using the nonlinear normalized energy index. Finally, experimental and simulation results are provided to compare the proposed approach with existing methods and verify its effectiveness.

In this article, the zonotopic set-membership state estimation (ZSMSE) for cyber-physical systems (CPSs) is investigated in the presence of stealthy false data injection (FDI) sensor attacks. A set-membership state estimation approach is established in which the true state is enclosed by a parallelotope (a special class of zonotope) at each sampling instant. Due to the openness of the network transmission process, malicious attackers have the ability to modify sensor measurement information by injecting false data. Therefore, from the perspective of the defender, the condition for an attack to bypass the detector and destroy the ZSMSE is discussed. Specifically, the stealthiness definition for the FDI sensor attacks and the vulnerability definition for ZSMSE are given. Moreover, the necessary and sufficient condition for the ZSMSE to be vulnerable is derived. Subsequently, a watermarking-based protection strategy is proposed for vulnerable ZSMSE to ensure the validity of state estimation results under stealthy attacks. The watermarking matrix is designed to break the stealthiness of the attack while ensuring the convergence of the ZSMSE. Finally, a series of illustrative examples are provided to demonstrate the effectiveness of the proposed strategy.

Under current energy development trends, remote clean energy is increasingly integrated into heavily loaded receiving-end power grids, replacing conventional synchronous generators. This shift leads to grid hollowing and reduced voltage support capability. Concurrently, with a high proportion of induction motor load equipment integrated into grid, power system faces a high risk of transient voltage instability caused by dynamic load instability. To guide the grid and equipment with the ability to ride through transient voltage excursions, grid codes formulate transient voltage recovery criterion (TVRC). Directly targeting TVRC compliance, this paper proposes a reinforcement-learning (RL)-based adaptive load shedding strategy with an offline centralized training & real-time decentralized execution framework. Firstly, the load shedding control is formulated as a Markov decision process (MDP), where states are defined by TVRC-based transient voltage trajectories, and actions include load shedding amounts and time delays. Reward functions are designed according to control goals and control costs. Secondly, a load-shedding agent and a time-delay agent represented by deep Q network (DQN) are designed, and corresponding target DQNs are further introduced to construct a double double-DQN (D2DQN) training structure. Agents learn optimal control policies through offline centralized training using collected state-action-reward sequences from multiple load buses. Subsequently, a decentralized load shedding strategy is formed with the well-trained agents deployed at load shedding buses, enabling real-time adaptive determination of shedding locations, amounts, timing, and rounds according to transient voltage states. This approach makes voltage recovery by satisfying the TVRC with the lowest cost of load shedding. Simulations on the Nordic test system and a provincial receiving-end power system validate the effectiveness and adaptability of the proposed strategy.

This paper develops an entropy-based stability and robustness framework for nonlinear hypergraph dynamics with conservation and flow balance. We consider generator-form systems on the simplex whose state-dependent transition rates capture higher-order (tensor) interactions among nodes. Under a tensor generalized detailed-balance (TGDB) condition, we show that the system admits a unique equilibrium and an entropy Lyapunov function ensuring global asymptotic stability. The Jacobian restricted to the tangent subspace of the simplex is Hurwitz, and its spectral gap determines the exponential convergence rate. Building on this structure, we derive first-order sensitivity bounds of the equilibrium under perturbations of the coupling tensor and establish a local input-to-state stability (ISS) estimate with respect to external inputs. The results reveal a quantitative link between the spectral gap and the system's robustness margin: larger spectral gaps imply smaller equilibrium shifts and faster recovery under structural or parametric perturbations. Numerical experiments on tensor-coupled flow models confirm the theoretical predictions and illustrate how the proposed entropy-dissipating framework unifies stability and robustness analysis for conservative higher-order network systems.

This work extends the distributed Nash equilibrium seeking algorithm to address multiagent games under time-varying, signed, and undirected communication networks. Initially, based on uniform complete observability for linear time-varying systems, a new analytical framework is developed to determine the necessary and sufficient condition for achieving globally uniformly exponential stability of the error dynamics of the distributed estimator. It is shown that exponential stability is realized if and only if the joint (is an element of, T)-connectivity and the negative-link assumption of time-varying signed graphs are satisfied. Subsequently, it is demonstrated that under some mild conditions, the seeking algorithm ensures the global uniform exponential convergence of players' strategies toward the Nash equilibrium.

Stochastic and high-power fluctuations of large-scale photovoltaic generations in distribution networks lead to complex power flow variations and voltage violations, posing significant challenges to voltage control. To address these challenges, this paper puts forward a knowledge-data driven centralized-decentralized coordinated four-step voltage control strategy to effectively dispatch heterogeneous voltage regulation devices. Step 1 proposes an optimal power flow model to determine the day-ahead voltage control results by regulating the taps of the on-load tap changer, the number of capacitor banks, and the charging/discharging power of battery energy storage systems, thereby minimizing daily network loss and preventing slow-time-scale voltage violations. Step 2 generates the voltage-regulation dataset through power flow and volt/var optimization calculations, establishing the data foundation for data-driven learning. Step 3 develops an intelligent inverter-based voltage controller by using fuzzy control theory for photovoltaic generations and battery energy storage systems, with voltage regulation knowledge embedded. Furthermore, a data-driven gradient descent learning method is presented for controller parameter optimization, enhancing global voltage regulation performance. Step 4 forms an online decentralized voltage control strategy with optimized voltage controllers to perform effective reactive power control adaptively according to operation states, thereby addressing frequent voltage violations and optimizing network power loss. Simulation results based on the IEEE33-bus system and a large-scale Caracas 141-bus system show that the proposed strategy can effectively maintain bus voltages within a secure range and reduce the network power loss by approximately 49% and 37%, respectively for the two systems, thereby validating its effectiveness and superiority.

This article investigates the effect of irrational behaviors on the expected convergence speed to the Nash equilibrium (NE) set within the log-linear learning (LLL) model. First, a significant augmentation for the classical LLL model is proposed by reconstructing the strategy update probability, merged with the effect of irrational behaviors. By incorporating such irrationality, it proves that under certain conditions, underweighting probabilities improves the expected convergence speed while overweighting probabilities reduces it. Second, these new results are applied to discuss the expected convergence speed to the stochastically stable Nash equilibrium (SSNE) set in potential games with a typical irrationality model, namely, probability sensitivity. Finally, the effectiveness of the proposed methods is illustrated by a sensor deployment example.