• • The TD3-R_LADRC strategy reduces DC bus voltage fluctuations by optimizing both observer and controller bandwidths via TD3, achieving faster convergence than fixed-parameter LADRC; this directly mitigates the risk of converter shutdown or reduced battery lifetime in microgrids with high renewable penetration.
• • The improved LESO estimates the disturbance derivative and applies order reduction to known states, enhancing observation accuracy without increasing system order; this lowers computational burden and parameter tuning complexity, enabling practical deployment on embedded controllers with limited resources.
• • Comparative experiments under renewable intermittency and load steps show TD3-R_LADRC outperforms dual-loop PI and conventional LADRC in disturbance rejection and robustness, with reduced settling time and overshoot; this translates to higher power quality and fewer protection trips in commercial energy storage systems.
• • The TD3 algorithm addresses Q-value overestimation and local optima issues inherent in DDPG, as evidenced by stable training convergence and improved control performance; this provides a reliable reinforcement learning framework for online parameter adaptation in safety-critical power electronics.