Deep Deterministic Policy Gradient-Based Actor–Critic Reinforcement Learning for Torque Ripple Minimization in Switched Reluctance Motors

The aim of this research is to investigate and reduce the torque ripple in Switched Reluctance Motor (SRM) drives, which is one of the major barriers to their acceptance for electric vehicle propulsion applications despite the advantages of robustness, efficiency, and wide operating range. High torque ripple not only deteriorates drive smoothness but also contributes to noise and vibration, demanding an advanced control strategy beyond traditional current-shaping and switching-based approaches. In this context, this work proposes a DDPG (Deep Deterministic Policy Gradient) Actor–Critic Neural Network-based reinforcement learning control framework that learns the optimal firing angle offsets dynamically to ensure less ripple electromagnetic torque under varying speeds and load conditions. The developed strategy has been designed and trained in MATLAB Simulink R2024b and then deployed in real time using an FPGA-based digital controller for validation on hardware. Comparative analysis with TSF (Torque Sharing Function) and DITC (Direct Instantaneous Torque Control) demonstrates that the reinforcement learning approach gives a much smoother torque response with better dynamic behavior over the operating range analyzed.