Enhanced Multiagent Reinforcement-Learning-Aided Adaptive Fractional-Order EADRC for Load Frequency Control of Multiarea Power Systems
提出一种自适应分数阶误差型自抗扰控制方法，结合多智能体强化学习在线调参，用于多区域电力系统负荷频率控制，显著降低频率偏差和误差积分。
High penetration of inverter-based renewables introduces pronounced volatility and heterogeneity to power systems. Coordinating heterogeneous frequency regulation units (FRUs) for robust load frequency control (LFC) remains challenging due to their disparate dynamics and capacities. To improve the frequency regulation capability of FRUs, an adaptive fractional-order error-based active disturbance rejection control (FO-EADRC) approach is proposed in this article. First, a fractional-order extended state observer (FO-ESO) is designed to reconstruct and compensate for the total disturbance of each FRU. Each FRU is equipped with an independent FO-EADRC controller, enabling the modular and independent regulation. Next, the closed-loop system stability based on FO-EADRC is analyzed via fractional-order (FO) theory. To further maximize the frequency regulation performance, a multiagent gated recurrent unit (GRU) soft actor-critic (SAC) algorithm is proposed to tune FO-EADRC gains online. Case studies are performed on a two-area power system. Quantitative results indicate that the proposed method yields maximum reductions of 64.5% in frequency deviation and 81.9% in the integral of absolute error compared with conventional methods, reducing the frequency deviation variance by 82.9% under ±50% parameter perturbations. These results demonstrate the effectiveness, significant superiority, and robustness of the adaptive FO-EADRC.