Premium
A Double‐Deep Q‐Network‐Based Energy Management Strategy for Hybrid Electric Vehicles under Variable Driving Cycles
Author(s) -
Zhang Jiaqi,
Jiao Xiaohong,
Yang Chao
Publication year - 2021
Publication title -
energy technology
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.91
H-Index - 44
eISSN - 2194-4296
pISSN - 2194-4288
DOI - 10.1002/ente.202000770
Subject(s) - artificial neural network , q learning , adaptability , reinforcement learning , variable (mathematics) , automotive engineering , energy management , computer science , curse of dimensionality , electric vehicle , driving cycle , engineering , artificial intelligence , energy (signal processing) , power (physics) , mathematics , ecology , mathematical analysis , statistics , physics , quantum mechanics , biology
As a core part of hybrid electric vehicles (HEVs), energy management strategy (EMS) directly affects the vehicle fuel‐saving performance by regulating energy flow between engine and battery. Currently, most studies on EMS are focused on buses or commuter private cars, whose driving cycles are relatively fixed. However, there is also a great demand for the EMS that adapts to variable driving cycles. The rise of machine learning, especially deep learning and reinforcement learning, provides a new opportunity for the design of EMS for HEVs. Motivated by this issue, herein, a double‐deep Q‐network (DDQN)‐based EMS for HEVs under variable driving cycles is proposed. The distance traveled of the driving cycle is creatively introduced as states into the DDQN‐based EMS of HEV. The relevant problem of “curse of dimensionality” caused by choosing too many states in the process of training is solved via the good generalization of deep neural network. For the problem of overestimation in model training, two different neural networks are designed for action selection and target value calculation, respectively. The effectiveness and adaptability to variable driving cycles of the proposed DDQN‐based EMS are verified by simulation comparison with Q‐learning‐based EMS and rule‐based EMS for improving fuel economy.