Optimising Damping Control in Renewable Energy Systems through Reinforcement Learning within Wide-Area Measurement Frameworks

This paper introduces a reinforcement learning-based controller, utilising the deep deterministic policy gradient (DDPG) method, to mitigate low-frequency disturbances in electrical grids with renewable energy sources. It features a novel reward function inversely related to the control error and employs a state vector comprising absolute and integral errors to enhance error reduction. The controller, tested on a dual-region system with solar power, utilises phasor measurement unit (PMU) data for global inputs. Its performance is validated through time-domain simulations, pole-zero mapping, modal analysis, frequency response, and participation factor mapping, using a custom MATLAB and Simulink toolkit. The design accounts for communication delays and adapts to variable conditions, which proves to be effective in reducing oscillations and improving system stability.