Machine Learning Applications in Fleet Electrification: Optimizing Vehicle Maintenance and Energy Consumption

Abstract

The electric vehicle (EV) market has been growing not only for passenger vehicles, where we witness high adoption of these cleaner cars, divided into Hybrid Electric Vehicles (HEVs), Plug-In Electric Vehicles (PHEVs), and Battery Electric Vehicles (BEVs), but also in buses, trucks, and vans, with an increasing conversion pressure due to regulations. In last-mile delivery, powerful optimization tools combining machine learning with operations research have become the subject of several studies, aiming to achieve efficient energy consumption in EVs by predicting drivers’ behaviors or track charts. Moreover, new strategies have been developed for something known as fleet electrification, i.e., when a carrier company aiming mainly to decrease its environmental footprint combines the operation of conventional fuel-powered vehicles with electric ones; this is not a complete switchover, but an intermediate stage.

Since computational methods converged, fleet electrification attracted researchers’ interest as a techno-economical optimization question related to three main research lines with a strong potential impact on environmental electricity: energy mix-related carbon emissions. The first category of services that impacted fleet electrification before the advent of machine learning was related to condition-based maintenance. As commercial gas, electricity, heating oil, and water consumption can be budgeted across a year’s worth of monthly payments, fixing a price through the year could help these fleets include commercial formulations that account for released corporate emissions through approved carbon offsetting. In concrete numbers, our technical paper has two key goals: firstly, we investigate whether and to what extent the Vehicle Misuse Factor, as defined, and the initial state of relevant vehicle parts (tires, gearbox, motor, etc.) affect the electric energy consumption of EVs; secondly, we propose and solve the cost minimization model arising from this energy consumption prediction.