Cloud-based health-conscious energy management of hybrid battery systems in electric vehicles with deep reinforcement learning. (1st July 2021)

• Record Type: Journal Article
• Authors: Li, Weihan; Cui, Han; Nemeth, Thomas; Jansen, Jonathan; Ünlübayir, Cem; Wei, Zhongbao; Feng, Xuning; Han, Xuebing; Ouyang, Minggao; Dai, Haifeng; Wei, Xuezhe; Sauer, Dirk Uwe
• Is Part Of: Applied energy. Volume 293(2021)
• Publication Date: 2021-07-01
• Subjects: Energy management -- Vehicle-to-cloud -- Reinforcement learning -- Battery aging -- Lithium-ion -- Battery safety; Power (Mechanics) -- Periodicals; Energy conservation -- Periodicals; Energy conversion -- Periodicals; 621.042
• DOI: 10.1016/j.apenergy.2021.116977
• Languages: English
• ISSNs: 0306-2619

Abstract

Abstract: In order to fulfill the energy and power demand of battery electric vehicles, a hybrid battery system with a high-energy and a high-power battery pack can be implemented as the energy source. This paper explores a cloud-based multi-objective energy management strategy for the hybrid architecture with a deep deterministic policy gradient, which increases the electrical and thermal safety, and meanwhile minimizes the system's energy loss and aging cost. In order to simulate the electro-thermal dynamics and aging behaviors of the batteries, models are built for both high-energy and high-power cells based on the characterization and aging tests. A cloud-based training approach is proposed for energy management with real-world vehicle data collected from various road conditions. Results show the improvement of electrical and thermal safety, as well as the reduction of energy loss and aging cost of the whole system with the proposed strategy based on the collected real-world driving data. Furthermore, processor-in-the-loop tests verify that the proposed strategy can achieve a much higher convergence rate and a better performance in terms of the minimization of both energy loss and aging cost compared with state-of-the-art learning-based strategies. Graphical abstract: Highlights: Cloud-based health-conscious energy management for hybrid battery systems. Power-split considering electro-thermal safety, energy efficiency and aging cost. DDPG-based strategy trained and validated with real-world vehicle data. Processor-in-the-loop tests with a machine learning-capable embedded device. Outperforming other approaches in training efficiency and system performance.