TD3 Deep Reinforcement Learning Control Approach for a DC‐DC Multiport Converter Used in Low‐Power EV Applications

ABSTRACT This paper presents a deep reinforcement learning (DRL)‐based control framework for a bidirectional dual‐input single‐output (BDISO) converter applied to low‐power electric vehicles (EVs). The study addresses the dynamic instability introduced by constant power loads (CPLs), which emulate the negative impedance behaviour of EV motor drives. A twin delayed deep deterministic policy gradient (TD3) algorithm is employed to directly generate control signals, thereby eliminating reliance on explicit converter models and intermediate control layers. The proposed model‐free approach enables fast voltage regulation and stable bidirectional power flow between sources and load, particularly under regenerative operating conditions. Simulation results obtained in MATLAB/Simulink demonstrate that the proposed strategy achieves up to 4–7 times faster settling time and 30%–90% lower voltage error compared with conventional proportional‐integral (PI) and super‐twisting algorithm (STA) controllers. These findings establish DRL‐based TD3 control as a scalable, model‐free solution for robust and efficient energy management in next‐generation EV powertrains.