Energy Consumption Prediction with Explainable AI

Proposal for a data-based methodology

Abstract

To reduce emissions in road traffic, the battery-electric drivetrain is gaining importance, not least due to its high efficiency. Therefore, the optimization of the onboard power supply system, which provides power to auxiliary consumers, is becoming a development focus. The share of their total consumption is measurably significant for overall efficiency due to the steadily increasing number of comfort functions in the vehicle. Since the energy supply of the auxiliary consumers in the onboard power system, together with the communicative networking of the components, forms a complex system, obtaining the necessary system understanding for optimization is not trivial. Hence, a data-based methodology is proposed, with which the power consumption behavior of individual consumers in the onboard power system is modeled based on their communication with other consumers. Subsequently, the models are validated using selected methods of explainable artificial intelligence (XAI). Additionally, a study on the application of individual global and local XAI methods, conducted with potential users in the relevant field of an automobile manufacturer, is presented and interpreted.