Predicting Vehicle Usage Using Incremental Machine Learning

Abstract

Today, there is an ongoing transition to more sustainable transportation, and an essential part of this transition is the switch from combustion engine vehicles to battery electric vehicles (BEVs). BEVs have many advantages from a sustainable perspective, but issues such as limited driving range and long recharge times slows down the transition from combustion engines. One way to mitigate these issues is by performing battery thermal preconditioning, which increases the energy efficiency of the battery. However, to optimally perform battery thermal preconditioning, the vehicle usage pattern needs to be known, i.e., how and when the vehicle will be used. This study attempts to predict the departure time and distance of upcoming drives using different incremental machine learning models. Further, the study includes a sensitivity analysis investigating how sensitive the performance of the battery thermal preconditioning is to incorrect predictions. The problem of predicting departure time and trip distance is approached in two different ways. The first approach only considers the first drive each day, while the second approach considers all drives that directly follows a charging session. The incremental machine learning models are trained and evaluated on historical driving data collected from a fleet of BEVs. Additionally, the prediction models are extended to quantify the uncertainty of their predictions, which can be used as guidance to whether the prediction should be used or dismissed. The best-performing prediction models yield an aggregated mean absolute error of 2.6 hours when predicting departure time and 12.5 km when predicting trip distance. However, for the considered temperatures and distances, the sensitivity analysis shows that the battery thermal preconditioning requires more precise predictions. The performance of the battery thermal preconditioning is sensitive, as the energy that can be saved from accurate predictions is less than what may be lost by adapting the preconditioning to incorrect predictions.