Ensemble-Based Robust Model for Accurate Driving Range Estimation of EVs Leveraging Big Data

In the emergence of greener transportation, electric vehicles (EVs) play an important role, where the accurate pre-diction of the driving range is pivotal for alleviating driver range anxiety, serving as a foundation for spatial planning, operational strategies, and efficient charging infrastructure management. This study addresses the challenge of limited driving range in EVs and introduces an ensemble-based machine learning (ML) model for predicting the vehicle's driving range. By employing big data collected from the real world, this work investigates the appropriateness of various ML models, including extreme learning model (ELM), extreme gradient boosting (XGBoost), multiple linear regression (MLR), multilayer perceptron (MLP), deep MLP, random forests (RF), AdaBoost, and support vector regression (SVR) for the problem stated above. Extensive experimental analysis proves that the ensemble framework that exploits MLR and XGBoost predictors surmounts the existing solutions. It achieves an R2 score greater than 0.9 on both training and testing data subsets, exhibiting no overfitting issues, and boasting acceptable inference time. The findings offer a compelling solution for enhancing the estimation of EV driving range for practical applications.

I. Introduction

According to the International Energy Agency (IEA), the application of EVs in global transportation is projected to grow rapidly, promising a noteworthy reduction of carbon emissions by 45 percent from 2010, and a complete elimination by 2050 [1]. This impending shift is poised to yield significant benefits for power grid economics, enhancing resilience, relia-bility, and sustainability [2]. In response to this trend, numerous countries, including Canada and the UK, have not only declared but also implemented policies aimed at expediting the transition from Internal Combustion Engine (ICE) vehicles to EVs by 2040. Conversely, countries, like China have already taken the bold step of halting new investments in ICE production facili-ties since 2019, thereby facilitating the proliferation of EVs. Additionally, in the United States, approximately ten states have enforced regulations mandating zero-emission vehicles, with California alone committing to deploy five million EVs on its roads by 2030. As per estimations, the EV count is projected to exceed 250 million by 2030, with a corresponding rise in the total electricity demand, anticipated to reach 1.1 PWh. Although EVs are becoming more popular, the greatest barriers preventing the widespread use of EVs are range limitation, long charging times, and poor accessibility to charging stations (CSs) [3]. As a result, many consumers are still hesitant to buy a fully battery-powered EV, although the governments provide several incentives to encourage people to acquire EVs [4].