ISBN 978-3-8439-4942-2

978-3-8439-4942-2, Reihe Informatik

Andreas Heimrath
An approach to a machine learning-based operating strategy in automotive electrical energy management

197 Seiten, Dissertation Technische Universität München (2021), Softcover, B5

Zusammenfassung / Abstract

Vehicles have to reduce their CO2 emissions and increase their energy efficiency to meet future regulatory requirements. However, current approaches to realize operating strategies for automotive electrical energy management such as rule-based operating strategies are a hurdle for the integration of additional signals to enhance the energy efficiency. Furthermore, they suffer from a low level of automation during the product development, while the complexity of the electrical system and the number of product variants are increasing. In this thesis, reinforcement learning (RL) is identified to overcome these limitations. Therefore, this thesis suggests the concept of a reinforcement learning-based operating strategy to contribute to enhancing the energy efficiency of a 12V electrical system. The concept was evaluated in the simulation of an electrical system as well as in a real vehicle during 23.5 h of decision-making under real driving conditions. The suggested concept of a RL-based operating strategy has the potential to replace rule-based operating strategies and is a foundation for future studies aiming at maximizing its contribution to energy efficiency beyond the methodological focus of this thesis.