Xiaowei Zhao

Wind energy has expanded rapidly in recent years, leading to increasingly dense clusters of offshore wind farms. As a result, wind farm wake effects, manifested as reduced wind speeds downstream of operating turbines, have become an important consideration for wind resource assessment and large-scale planning. Here we investigate near-surface wake signatures using wind speeds retrieved from 7122 Sentinel-1A/B synthetic aperture radar (SAR) images acquired between 2020 and 2022, covering more than 60 offshore wind farms across Europe and Asia. A consistent processing workflow and an automated wake-detection algorithm are applied to identify wake-affected regions and quantify wake-related wind speed changes at 10 m height. The results show that near-surface wake signatures can persist over distances exceeding 100 km under favorable conditions and that wake-affected regions exhibit an average wind speed reduction of 0.990 m/s (12.4%) at 10 m height. Several cases of wake interactions extending across national boundaries are also observed in densely developed offshore regions. These findings provide a large-scale, observation-based characterization of near-surface wind farm wake signatures.

This paper develops a prescribed performance-based reinforcement learning (RL) hybrid resilient and fault-tolerant control scheme for wind turbine torque regulation. Focusing on ensuring optimal rotor speed performance in maximum power point tracking (MPPT) control under simultaneous faults and attacks, a finite-time prescribed performance function (PPF) is integrated with RL actor–critic networks. Distinct from mainstream RL-based control methods, the proposed approach compensates for blocked measurement information via a compensator-based state observer, followed by a neural network to approximate unknown nonlinearities induced by faults and attacks. Furthermore, actor–critic networks are employed to approximate the optimal control policy by solving a Hamilton–Jacobi–Bellman equation involving long-term prescribed tracking performance constraints. This design enables both fault-tolerant and attack-resilient capabilities, enhancing the reliability of power generation control. Case studies on the OpenFAST simulator indicate the robust MPPT performance under complex scenarios involving simultaneous faults and attacks. Specifically, under torque actuator faults, the proposed method reduces the tracking error by approximately 84% and 86% under false data injection (FDI) and denial-of-service (DoS) attacks, respectively, compared to the baseline controller, and further outperforms the robust PPF-based sliding mode control by about 37% and 27%.

Direct current microgrids (DCmGs) are attractive due to their high efficiency and ease of deployment, yet decentralized controllers remain vulnerable to false-data-injection (FDI) attacks and abrupt load changes. This article develops a resilient decentralized control architecture that achieves performance guarantees under bounded FDI attacks using only local measurements. The key idea is a decentralized high-gain observer that reconstructs each power buffer input impedance and stored energy while explicitly accounting for network coupling. On this basis, a smooth nonlinear feedback term provides attack compensation without chattering. In contrast to existing schemes, our design comes with explicit linear matrix inequality conditions that, first, certify asymptotic stability in the attack-free case and, second, ensure uniform ultimate boundedness of the closed loop under FDI, thereby yielding transparent tuning ranges for the controller and observer gains. Hardware-in-the-loop experiments, in which the DCmG is subjected to staged FDI attack windows and load steps, are conducted to demonstrate the proposed design with robust and practically tunable resilience for DCmGs.

This article presents a robust adaptive tracking control scheme for a class of nonlinear strict‐feedback systems. Building upon robust control principles for disturbance attenuation and the nonlinear backstepping technique for stabilization, the proposed approach unifies robust tracking design with online adaptive dynamic programming for strict‐feedback nonlinear systems. Finite‐energy disturbances are explicitly modeled as an adversarial player in a zero‐sum game, thereby bridging the gap between robust tracking synthesis and data‐driven online optimization. To approximate the Nash equilibrium, a novel Actor‐Critic‐Disturbance learning algorithm is developed, where the sufficiency of excitation condition combined with the experience replay mechanism ensures adaptive weight convergence without relying on the restrictive persistence of excitation. Furthermore, the value function is decomposed into a quadratic part and a nonlinear component, thereby guaranteeing satisfactory tracking performance while avoiding direct solutions of the Hamilton–Jacobi–Isaacs equation. Theoretical analysis establishes rigorous justification for the stability of the overall closed‐loop system, and simulation studies validate the effectiveness of the proposed method.

Large language models (LLMs) have revolutionized text-based code automation, but their potential in graph-oriented engineering workflows remains under-explored. We introduce SimuAgent, an LLM-powered modeling and simulation agent tailored for Simulink. SimuAgent replaces verbose XML with a concise, dictionary-style Python representation, dramatically cutting token counts, improving interpretability, and enabling fast, in-process simulation. A lightweight plan-execute architecture, trained in two stages, equips the agent with both low-level tool skills and high-level design reasoning. To tackle sparse rewards in long-horizon tasks, we propose Reflection-GRPO (ReGRPO), which augments Group Relative Policy Optimization (GRPO) with self-reflection traces that supply rich intermediate feedback, accelerating convergence and boosting robustness. Experiments on SimuBench, our newly released benchmark comprising 5300 multi-domain modeling tasks, show that a Qwen2.5-7B model fine-tuned with SimuAgent converges faster and achieves higher modeling accuracy than standard RL baselines, and even surpasses GPT-4o when evaluated with few-shot prompting on the same benchmark. Ablations confirm that the two-stage curriculum and abstract-reconstruct data augmentation further enhance generalization. SimuAgent trains and runs entirely on-premise with modest hardware, delivering a privacy-preserving, cost-effective solution for industrial model-driven engineering. SimuAgent bridges the gap between LLMs and graphical modeling environments, offering a practical solution for AI-assisted engineering design in industrial settings.