Driving Range Estimation of Electric Vehicles using Deep Learning
George Dona, P. Sivraj
Abstract
Electric vehicles (EV) are gaining popularity due to their reduced pollution, fewer emissions, and energy savings but has big challenges like driver's range anxiety, that slows down the penetration of electric vehicles. This work focuses on analyzing the impact of different factors affecting the driving range of EVs and developing a deep learning-based optimized system for range prediction using neural networks. An electric vehicle model is developed in MATLAB Simulink to analyze the impact of various factors on the driving range and validation of the range prediction model. A comparative study on different neural network models is done using the Keras framework to select the best suited regression model to predict the driving range based on prediction error. The analysis reveals the impact of various factors like driver behaviour, exploitation environment, battery parameters, and auxiliary loads on the driving range. For the test data, the bidirectional long short-term memory shows the minimum error compared to other sequential models with a mean square error of 0.029 km in the prediction of driving range. The work can be further extended by integrating to driver assistance systems for best route selection, charge scheduling, trip planning etc.