On the Use of Machine Learning for State-of- Charge Forecasting in Electric Vehicles

Nowadays, it is well known that a main solution for pollution reduction in cities will be the introduction of electric and hybrid vehicles on transportation roads. Many research efforts have been dedicated to develop new technologies to further promote the use of this type of vehicles. However, their penetration on transportation roads faces some obstacles that have not yet been fully tackled. For instance, the development of intelligent battery management systems needs to be further investigated taking into consideration the uncertainty linked to how vehicles will perform in different scenarios, such as traffic situation, driver behavior, and road profile. The work presented in this study is towards developing a battery management system by investigating new approaches for accurate estimation and prediction of remaining charge, the expected lifetime of the batteries and the remaining driving rang. We focus mainly on the integration of predictive analytics techniques for forecasting the state-of-charge. We first deployed statistical- and machine learning based techniques in real-sitting scenarios (LSTM, ARIMA and XGBoost). Experiments have been conducted using an electric vehicle platform and results are reported to shed more light on their accuracy for multiple-horizon forecasts of battery's state of charge.