Multi-objective optimization of main transmission parameters in integrated flywheel drive system based on improved SAC algorithm

This paper proposes a multi-objective optimization method based on the Stein Soft Actor-Critic algorithm for determining the optimal transmission parameters of an integrated flywheel drive system. A deep neural network surrogate model is trained on 216 orthogonal experimental design samples to replace conventional co-simulation, thereby accelerating the training process. Meanwhile, Stein variational gradient descent is embedded into the Soft Actor-Critic policy network, enabling particle-based policy optimization and enhancing exploration capability in the continuous action space. Experimental results show that, compared with Soft Actor-Critic, Twin Delayed Deep Deterministic Policy Gradient, Deep Deterministic Policy Gradient, and Multi-Objective Particle Swarm Optimization, the proposed method achieves improvements in both state of charge and driving range under standard driving cycles, and is validated on a real-world driving route, demonstrating its effectiveness and generalizability.