Size and control co-optimization of electric vehicles using Bayesian optimization

The optimal design of electric vehicle (EV) powertrains is essential for high energy efficiency. Since traditional design approaches rely on deterministic driving cycles, the design overlooks the stochastic properties of real-world traffic and hence leads to non-robust results. To enhance robustness, stochastic design and optimization methods have been introduced to explicitly account for driving uncertainty by minimizing the expected energy consumption over stochastic driving scenarios. Since stochastic design methods require optimization for many random driving trajectories, they are computationally prohibitive. This paper proposes an efficient bi-level stochastic co-optimization method for the robust design of a typical EV powertrain under stochastic driving conditions. The outer layer searches for the optimal and robust powertrain size via Bayesian Optimization (BO) and an adaptive sampling (AS) method. The inner layer employs optimal energy management control to find the greatest energy efficiency for the size proposed by the outer layer. Compared with a population-based Genetic Algorithm combined with adaptive sampling, the proposed BO-AS method achieves a comparable optimal energy efficiency but reduces the computation time by 93%. Compared to a fixed-size Monte Carlo sampling method, BO-AS achieves equivalent energy efficiency with a 60–75% reduction in computation time. The results demonstrate that integrating adaptive stochastic evaluation with sample-efficient BO provides a practical and scalable solution for uncertainty-resilient EV powertrain design. Compared to a baseline EV, BO-AS achieves approximately 3% and 70% reductions in mean and variance of energy consumption. The new design reduces vehicle mass by 1.4% and power-to-mass ratio by 5.6%. Although the battery capacity is reduced by 13.7%, the expected range is decreased by only 11%.