Designing ML-based Prediction Systems for the Energy Consumption of Electric Buses to Support Public Transport Providers

Abstract

The European Green Deal requires all public transport providers to transform their fleets to sustainable driving technologies, such as electric buses, by 2030. However, e-bus ranges are limited and dependent on numerous factors, complicating operations. While researchers are already examining how machine learning models can predict driving ranges, the literature lacks findings about how IT artifacts should be designed to support public transport providers in electric bus operations. Therefore, we examine this by applying the design science research paradigm and collaborating with a major public transport provider. We conducted three focus groups to formulate ten meta-requirements and derive four design principles. We instantiated these principles by developing a machine-learning model on real-world data and a web prototype, which fourteen experts evaluated. The evaluation approved the design principles and led to two further ones, including e-bus disposition and remedy wear. Finally, we discuss further research's impact, limitations, and demands.