Reinforcement learning-aided co-optimization of powertrain sizing and control for electric vehicles

The integrated optimization of powertrain control strategies and component parameters is a crucial approach to improving energy efficiency in electric vehicles (EVs), thereby enhancing its market competitiveness. However, the potential of existing optimization methods is constrained by the convexity and dimensionality of the optimization problem. This paper proposes a reinforcement learning-aided integrated optimization method that leverages deep learning to predict the feasibility and optimal energy consumption of powertrain design schemes, significantly reducing the computational effort. Furthermore, it transforms the static parameter optimization problem into a dynamic trajectory optimization problem, improving both the generalizability capability and the optimization efficiency of the powertrain optimization process. This paper employs the Deep Deterministic Policy Gradient (DDPG) algorithm to iteratively search for the optimal design scheme, achieving a 6.91% reduction in energy consumption compared with the benchmark scheme. The study demonstrates the effectiveness of reinforcement learning in optimizing EV powertrain parameters, offering a novel approach for addressing similar problems.