Electric Vehicle's Range and State of Charge Estimations using AutoML

This paper examines the potential of AutoML for predicting the range and State of Charge (SOC) of Electric Vehicles (EVs). Unlike traditional SOC estimation methods, such as Coulomb counting, Equivalent Circuit Models (ECM), or Machine Learning (ML)-based approaches, Range Estimation Algorithms (REA) consider route-specific factors to offer more precise battery depletion predictions. However, ML-based REAs can be complex and time-consuming to train, necessitating a deep understanding of Artificial Neural Networks (NN) architecture and optimization strategies. AutoML addresses this issue by automating the selection of the optimal NN architecture, hyper- parameters, and data preprocessing techniques, making it more accessible for those with limited expertise to develop effective ML models. Our study centers on constructing SOC estimation and range estimation models using the AutoML library AutoGluon, developed by Amazon Web Services (AWS). Our findings indicate that while SOC estimation alone has limitations in predicting an EV's remaining range, REAs are specifically designed to overcome this challenge by building on SOC estimation to accurately forecast the remaining distance.