Energy management strategy for electric vehicles based on deep Q -learning using Bayesian optimization
- PDF / 1,827,171 Bytes
- 15 Pages / 595.276 x 790.866 pts Page_size
- 95 Downloads / 237 Views
(0123456789().,-volV)(0123456789(). ,- volV)
EXTREME LEARNING MACHINE AND DEEP LEARNING NETWORKS
Energy management strategy for electric vehicles based on deep Q-learning using Bayesian optimization Huifang Kong1 • Jiapeng Yan1
•
Hai Wang2 • Lei Fan1
Received: 3 January 2019 / Accepted: 5 October 2019 Springer-Verlag London Ltd., part of Springer Nature 2019
Abstract In this paper, a deep Q-learning (DQL)-based energy management strategy (EMS) is designed for an electric vehicle. Firstly, the energy management problem is reformulated to satisfy the condition of employing DQL by considering the dynamics of the system. Then, to achieve the minimum of electricity consumption and the maximum of the battery lifetime, the DQL-based EMS is designed to properly split the power demand into two parts: one is supplied by the battery and the other by supercapacitor. In addition, a hyperparameter tuning method, Bayesian optimization (BO), is introduced to optimize the hyperparameter configuration for the DQL-based EMS. Simulations are conducted to validate the improvements brought by BO and the convergence of DQL algorithm equipped with tuned hyperparameters. Simulations are also carried out on both training dataset and the testing dataset to validate the optimality and the adaptability of the DQL-based EMS, where the developed EMS outperforms a previously published rule-based EMS in almost all the cases. Keywords Energy management strategy (EMS) Electric vehicle (EV) Deep Q-learning (DQL) Bayesian optimization (BO)
1 Introduction Development and deployment of electric vehicles (EVs) have gained tremendous momentum nowadays mainly due to concerns over petroleum shortages and environmental pollution [1]. Nevertheless, the batteries in the EVs suffer severe battery degradation under high-rate charge or discharge operation mode caused by frequent and peak power demands [2–4]. On the other hand, another available energy storage, supercapacitor, is characterized by high power density and exceptionally long cycle lifespan [5], so it is much more robust in handling peak power and current requirements. Therefore, hybrid energy storage systems (HESS), where battery serves as a persistent energy source and supercapacitor is employed to share power load, are widely adopted [6–9]. The introduction of an additional & Jiapeng Yan [email protected] 1
School of Electrical Engineering and Automation, Hefei University of Technology, Hefei 230009, China
2
College of Science, Health, Engineering and Education, Murdoch University, Perth, WA 6150, Australia
energy source increases the complexity of power flow; therefore, an energy management strategy (EMS) should be carried out to coordinate the power distribution between the two energy sources. Previously published EMSs can be generally classified into three categories: the rule-based EMSs (such as the deterministic rule-based control strategies [10] and fuzzy logic control strategies [11]), the optimization-based EMSs (such as the dynamic programming algorithm [12] and
Data Loading...
