Learning a map to predict recuperation for efficiency in automotive electrical energy management

Abstract:
Today’s operating strategies in automotive electrical energy management have to satisfy safety requirements and the driver’s needs at any point of time. However, due to their major impact on a vehicle’s efficiency and CO2 emissions, operating strategies for high efficiency cannot rely only on sensors capturing the state of the vehicle’s components. They also need a prediction of the future environment conditions influencing the energy system. Integrating the prediction of recuperation into decision-making according to operating strategies is a promising approach to enhance efficiency. Re-cuperation enables charging the battery of the vehicle by converting kinetic energy during deceleration. Therefore, we suggest a data-driven approach to learn a map for the prediction of recuperation from fleet data. The approach does not only make it possible to construct the map offline from existing data, but it also enables a way to update the map online during drives of a fleet’s vehicles. This data-driven approach represents the actual behavior of the energy system with temporal and spatial resolution. Therefore, it overcomes the limitations of existing methods based on topographic profiles or infrastructure maps to predict recuperation. Finally, the learned map improves the prediction of recuperation and paves the way for fully exploiting the potential of machine learning-based operating strategies in automotive electrical energy management.