基于强化学习的多端直流系统柔性控制用于海上风电场并网

Converter power sharing and DC voltage regulation are critical control objectives in voltage source converter-based multi-terminal high-voltage DC (VSC-MTDC) systems, particularly when incorporating massive offshore wind farms (OWFs). However, the inherent variability and intermittency of OWFs pose significant challenges to conventional model-based and optimization-based control methods, which often rely on accurate system modeling and extensive computation. Moreover, traditional heuristic optimization approaches are prone to local optima, limiting their ability to handle rapidly fluctuating wind power conditions effectively. Reinforcement Learning (RL), with its capability to adaptively optimize control policies through continuous interaction with dynamic environments, presents a promising alternative. This paper proposes a novel RL-based control strategy for DC voltage regulation and proportional power sharing, leveraging an adaptive droop control mechanism. The Proximal Policy Optimization (PPO) algorithm is employed to dynamically adjust local droop coefficients, benefiting from its robustness in continuous action spaces and its ability to mitigate the effects of OWF power fluctuations. To enhance optimization efficiency and accelerate convergence, a combined MTDC system pre-evaluation and PPO initialization method is introduced. The proposed RL-based control strategy is validated through dynamic simulations of a five-terminal MTDC grid in MATLAB/Simulink, demonstrating superior performance in handling various disturbances and contingencies associated with OWF integration.