Improved Linear Active Disturbance Rejection Control of Energy Storage Converters Based on the TD3 Algorithm
Output voltage fluctuations in DC microgrids arise from renewable generation intermittency, spatiotemporal load variations, and external disturbances. This study proposes a reconstructed linear active disturbance rejection control strategy (TD3-R_LADRC) that integrates a twin delayed deep deterministic policy gradient (TD3) algorithm to enhance the DC bus voltage stabilization capability of battery energy storage interface converters. The improved linear extended state observer (LESO) estimates the derivative of the total disturbance and applies order reduction to known state variables, achieving faster and more accurate tracking and compensation without increasing system order. Frequency-domain performance and stability analyses are conducted for the proposed strategy. The TD3 reinforcement learning algorithm then optimizes the observer bandwidth and controller bandwidth of the improved LADRC, enabling precise observation and rapid convergence. Digital simulations and low-power experiments compare the proposed TD3-R_LADRC against conventional LADRC and dual-loop PI control under various operating conditions. Results demonstrate that TD3-R_LADRC exhibits superior disturbance rejection, stability, and robustness against renewable output uncertainty, load fluctuations, and external disturbances, effectively improving frequency stability control and offering theoretical and engineering value for energy storage converter applications.