Vergleich datengetriebener Kalibrationsmethoden für das Thermomanagement eines elektrischen Sportwagens

Autoren

P. Muhl – Porsche AG, T. Rudolf - Porsche Engineering GmbH, S. Hohmann - Karlsruher Institute of Technology, L. Eckstein - ika, RWTH Aachen University

Zusammenfassung

The quality of the thermal management (TMM) strongly influences the energy consumption and therefore long-range driving, especially in high-performance electric sports cars. The related control systems are integral parts of the energy management and vehicle operating strategies. Therefore, they need to be adapted to the full vehicle architecture as well as external disturbances. Though, design targets may conflict and result in a trade-off between efficiency versus overall performance. The performance of the controlled system is optimized by the calibration of the variable parameters. Closing the reality gap between simulation and the real-world application is challenging with conventional methods. This work presents data-based approaches to calibrate a parameter-variant volume flow controller for the TMM of a real car. Based on a parameter dataset, control performances of multi-dimensional parameter sets are analyzed. We study the approach of data-driven calibration with reinforcement learning. Promising parameter sets are derived and improved along a sequence of predictions based on real driving trajectories. We assess the results on controller performance criteria and evaluate the most promising parameter sets in the real-world system on a TMM test bench and based on a real-world high performance test drive. The discussion comprises data-efficiency, advantages, and drawbacks of the individual approaches in comparison to state-of-the-art methods. We conclude with an outlook toward the integration of the proposed methodology into the development process.