Analysing the driving load on electric vehicles using unsupervised segmentation models as enabler to determine the time of battery replacement and assess driving mileage

The aim of this contribution is to study the driving load on electric vehicles through data mining methods. The battery pack is currently the major cost driver for these vehicles. Thus, battery durability is an important lever to economically establish electric mobility in the mass market. A segmentation modelling approach, based on unsupervised clustering algorithms, evaluates battery management system field data from BMW i3 vehicles such as state of charge, temperature and current in terms of their historical distribution. Clustering algorithms, due to their frequent use in the field of machine learning and information retrieval, are able to analyse big quantities of vehicle data with the aim to group vehicles with similar driving behaviour. The cluster analysis is further validated using the method of calculating respective silhouette coefficients to assess clustering performance and the influence of input parameters. The analysis of histograms concludes with the definition of the most common types of drivers worldwide. The vehicle clusters can further be correlated with battery ageing in order to find suitable 2nd life applications as part of stationary energy storage systems.