DB-ELM: A Dynamic Optimization Algorithm for Longitudinal Inverse Modeling in Electric Vehicles

To improve the adaptive capabilities of longitudinal dynamic inverse modeling in electric vehicles across diverse road conditions, this paper presents an advanced machine learning inverse model based on the Dynamic Bayesian Extreme Learning Machine (DB-ELM). The proposed method takes the current vehicle speed, target acceleration, and road slope as inputs for the inverse model, and outputs the corresponding pedal opening to adapt to the adaptive adjustment of pedal opening under dynamic driving conditions. To enhance the adaptability and robustness of the model, the training process incorporates dynamic Bayesian techniques along with a multi-start strategy to ensure comprehensive and robust model performance. This approach effectively addresses challenges such as high validation errors, significant cumulative validation discrepancies, large fluctuations in regularization parameters, and unstable test errors, all while maintaining the original training speed. By ensuring robust adaptive performance, the algorithm remains highly responsive and effective under varying conditions. Validation of the proposed inverse model algorithm was conducted through the implementation of acceleration tests on a co-simulation platform, which were performed under varying road slopes.