Multi-objective virtual calibration method for parameters in two control strategies of a complex hybrid electric vehicle

Abstract

This paper presents a virtual calibration method for parameters in two coupled control strategy of a complex hybrid electric vehicle. Firstly, the sampling inputs are determined through the optimal Latin hypercube design, and the response are obtained by employing a validated physics-based model, forming the training dataset. Based on this dataset, the deep neural network model is employed to establish a surrogate model connecting control parameters and response, achieving a mean square error of 7.47 × 10⁻⁷. Using the deep neural network model, sensitivity analysis of calibration parameters is carried out through the Sobol indices method. The results of this analysis reveal calibration paths of key parameters and combinations, effectively reducing the search region. Finally, the double distance sorting genetic particle swarm optimization algorithm is employed to solve the multi-objective optimization problem. The optimization results indicate a 2.09 % reduction in the mean square error of speed, a 24.28 % reduction in equivalent fuel consumption, and a 14.71 % reduction in battery capacity loss compared to the initial values. These outcomes clearly confirm the effectiveness of the proposed calibration method in enhancing calibration efficiency and provide valuable theoretical guidance for actual calibration.