Adaptive control of electric vehicle drives through neural network ensembles

Abstract

This study examines the use of neural network ensembles in adaptive control for electric vehicle (EV) propulsion systems, using simulated data to evaluate their efficacy. The research aims to evaluate the collective performance of a group, analyze the characteristics of electric vehicle drives, examine the feedback from adaptive control systems, and analyze the data used to train neural networks in order to get a thorough understanding of the subject. The results demonstrate the resilience of neural network ensembles in predictive modeling, with mean squared error values ranging from 0.0028 to 0.0042 and R-squared scores between 0.979 and 0.992. An examination of electric vehicle (EV) driving characteristics reveals differences in battery capacity (ranging from 60 to 85 kWh) and motor efficiency (ranging from 85% to 95%). Notably, there are correlations that demonstrate the influence of weight on the needs for battery capacity. An analysis of the feedback parameters in adaptive control reveals speed inaccuracies ranging from -1.8 to -3.2 km/h, battery voltage errors between 1.5 and 2.8 V, temperature mistakes ranging from 1.2 to 2.5°C, and variations in the control signal. This highlights the significant impact these factors have on the adjustments made by the control system. Moreover, examination of the training data for neural networks emphasizes the significance of having a wide range of inputs (0.3-0.9) and the intricate connections between inputs and outputs (0.6-0.95). In summary, these findings highlight the ability of neural network ensembles to improve predictive accuracy, comprehend the dynamics of EV systems, and emphasize the importance of accurate feedback and high-quality training data for effective adaptive control strategies in electric vehicles. These insights are valuable for advancing EV technology and control methodologies.