Multi-physics coupled simulation and surrogate-assisted tri-objective optimization of electric sport utility vehicle performance across ambient temperatures

Abstract

In this study, an integrated framework that combines energyflow experimentation, multiphysics simulation, datadriven surrogate modeling and multiobjective optimization was developed to enhance the comprehensive performance of an electric sportutility vehicle (SUV). A high-resolution test bench was developed, and a coupled multi-physics simulation model was constructed and parameterized. Validation against experimental measurements showed that the relative errors of key variables, including vehicle speed, battery state of charge (SOC) and motor torque, were all below 3%, demonstrating excellent agreement between the model predictions and experimental results under diverse operating conditions. The calibrated model was embedded within an automated toolchain that enabled closedloop parameter ingestion, parallel simulation and result extraction, thereby markedly reducing computational overhead. EXtreme Gradient Boosting (XGBoost) was employed as the surrogate predictor which configured with 40 estimators and a learning rate of 0.23, and RMSE, R² and MAE on the test set reached 0.937, 0.967 and 0.699, demonstrating strong predictive accuracy and robust generalization capability. A triobjective optimization problem was formulated to maximize battery recovery energy (E _rec ), maximize halfshaft effective output power (W _eff ) and minimize energy consumption per 100 km (E _com ) at −7°C, 23°C and 35°C, solving by the MultiObjective Pelican Optimization Algorithm (MOPOA). The problem yielded 123 nondominated solutions, and representative candidates simultaneously improved E _rec , W _eff and E _com by 18.1%, 5.2% and 5.1%, with the best aggregate enhancement reaching 28.4 %. The proposed methodology provides theoretical guidance for design refinement and control calibration of electric SUVs under complex thermal conditions.