Scholarlink

<p>This study proposes a hybrid optimization and control framework for wind farm operation that integrates Quantum-Inspired Multi-Agent Reinforcement Learning (QI-MARL) with Non-dominated Sorting Genetic Algorithm III (NSGA-III) within a federated constraint negotiation (FCN) architecture. The framework is designed to address the coupled challenges of wake-induced aerodynamic interactions, stochastic atmospheric turbulence, and noise-regulation constraints through decentralized, cooperative turbine-level decision-making. The proposed methodology simultaneously optimizes inherently competing objectives (maximization of aggregate power production, reduction of structural fatigue loading, and mitigation of acoustic emissions) while strictly enforcing operational and regulatory constraints. Wake interactions are modeled using the Bastankhah Gaussian wake formulation, and the approach is evaluated on a five-turbine array of utility-scale 5-MW horizontal-axis wind turbines. Performance is assessed over 30 independent stochastic wind realizations, incorporating variability in wind speed and turbulence intensity. Relative to established control and optimization baselines (including greedy axial-induction control, conventional yaw-based wake steering, plain multi-agent reinforcement learning without quantum-inspired encoding, and NSGA-II-based multi-objective optimization) the proposed framework achieves an average 7.8% increase in total wind-farm power output, a 18.6% reduction in maximum blade-root damage-equivalent loads, and a 3.4 dBA decrease in far-field acoustic levels at the designated receptor location. In addition, Pareto-front quality is significantly improved, with a 22% increase in hypervolume and a 27% reduction in inverted generational distance relative to NSGA-III alone. These results demonstrate that the proposed quantum-inspired, federated multi-agent framework provides a robust, scalable, and wake-aware control paradigm capable of delivering fatigue-conscious, noise-compliant, and resilient wind-farm operation under complex and uncertain atmospheric conditions, thereby supporting the deployment of intelligent control strategies in next-generation sustainable wind energy systems.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Quantum-inspired multi-agent reinforcement learning and evolutionary optimization for stochastic wind farm control

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年3月7日

查看全文
  • Khamiss Cheikh,
  • E L Mostapha Boudi,
  • Rabi Rabi,
  • Hamza Mokhliss

摘要

This study proposes a hybrid optimization and control framework for wind farm operation that integrates Quantum-Inspired Multi-Agent Reinforcement Learning (QI-MARL) with Non-dominated Sorting Genetic Algorithm III (NSGA-III) within a federated constraint negotiation (FCN) architecture. The framework is designed to address the coupled challenges of wake-induced aerodynamic interactions, stochastic atmospheric turbulence, and noise-regulation constraints through decentralized, cooperative turbine-level decision-making. The proposed methodology simultaneously optimizes inherently competing objectives (maximization of aggregate power production, reduction of structural fatigue loading, and mitigation of acoustic emissions) while strictly enforcing operational and regulatory constraints. Wake interactions are modeled using the Bastankhah Gaussian wake formulation, and the approach is evaluated on a five-turbine array of utility-scale 5-MW horizontal-axis wind turbines. Performance is assessed over 30 independent stochastic wind realizations, incorporating variability in wind speed and turbulence intensity. Relative to established control and optimization baselines (including greedy axial-induction control, conventional yaw-based wake steering, plain multi-agent reinforcement learning without quantum-inspired encoding, and NSGA-II-based multi-objective optimization) the proposed framework achieves an average 7.8% increase in total wind-farm power output, a 18.6% reduction in maximum blade-root damage-equivalent loads, and a 3.4 dBA decrease in far-field acoustic levels at the designated receptor location. In addition, Pareto-front quality is significantly improved, with a 22% increase in hypervolume and a 27% reduction in inverted generational distance relative to NSGA-III alone. These results demonstrate that the proposed quantum-inspired, federated multi-agent framework provides a robust, scalable, and wake-aware control paradigm capable of delivering fatigue-conscious, noise-compliant, and resilient wind-farm operation under complex and uncertain atmospheric conditions, thereby supporting the deployment of intelligent control strategies in next-generation sustainable wind energy systems.