Gaussian Process-Based Multi-Objective Optimization of Hybrid Electric Vehicle Energy Management Under Uncertainty

Abstract

Hybrid Electric Vehicle powertrain optimization faces uncertainty when driving conditions vary, when components age, and when modeling errors occur, which conventional optimization methods often neglect. This paper presents the uncertainty-aware multi-objective Bayesian optimization framework for the HEV powertrain design and the energy management, which is implemented in the MATLAB Simulink environment. The proposed methodology uses the Gaussian Process surrogate models, which capture nonlinear system behavior and quantify predictive uncertainty, where this supports data-efficient exploration of the design space. The Pareto-aware acquisition functions balance competing objectives, which include fuel consumption minimization, drivability improvement, as well as battery state-of-charge sustainability. The simulation results show that the proposed framework achieves robust Pareto-optimal solutions using fewer than 40 high-fidelity simulations, which reduces computational cost by more than 75 percent compared to evolutionary optimization methods. Under stochastic driving and component uncertainty, the optimized solutions reduce fuel consumption variability by about 35 percent and battery state-of-charge fluctuations by about 38 percent compared to deterministic optimization. Among the evaluated architectures, the series parallel HEV delivers the best trade-off between efficiency and robustness, which achieves mean fuel consumption of 4.36 L per 100 km with reduced performance dispersion. The key novelty of this work lies in the integration of the uncertainty-aware multi-objective Bayesian optimization with the MATLAB Simulink based HEV modeling, which provides practical and simulation-efficient framework for robust powertrain optimization under real-world operating uncertainty.