Reinforcement Learning-Based Inertia and Droop Control for Wind Farm Frequency Regulation
Yanchang Liang, Li Sun, Xiaowei Zhao
Abstract
As more and more wind turbines (WTs) are installed, there is an increasing interest in actively controlling their power output to meet power set-points and to participate in the frequency regulation for the utility grid. Conventional inertial and droop control loops use fixed gains, making it difficult to utilise the kinetic energy of WTs in a wind farm in a synergistic manner based on real-time information. In this paper, the fixed gains are modified to adaptive gains to improve frequency support performance and reduce the impact on mechanical structures. The cooperative frequency control problem for all WTs in a wind farm is modelled as a decentralised partially observable Markov decision process (Dec-POMDP) and solved using a multi-agent deep reinforcement learning (MADRL) algorithm. MATLAB/Simulink and FAST are run in connection to simulate the frequency response of a wind farm, where FAST simulates the mechanical part of WTs and Simulink simulates the electrical part. Simulation results show that the proposed method is effective in reducing frequency drops and the impact of frequency control on the mechanical structure.