Date of Award
5-2024
Document Type
Dissertation
Degree Name
Doctor of Philosophy (PhD)
Department
Automotive Engineering
Committee Chair/Advisor
Jiangfeng Zhang
Committee Member
Benjamin Lawler
Committee Member
Srikanth Pilla
Committee Member
Rajendra Singh
Committee Member
Zheyu Zhang
Abstract
The increasing concerns about transportation pollution and fossil fuel depletion motivate many studies on vehicle electrification and advanced energy-saving propulsion systems. When Comparing with traditional internal combustion engine vehicles, electrified vehicles, such as battery and supercapacitor electric vehicles, are equipped with more than one power source in the hybrid propulsion system, which can save more energy through efficient power combinations. Lithium-ion batteries are the preferred choice for energy storage in electric vehicles due to their superior energy density and cost-effectiveness. Nevertheless, matching the required power input and output leads to in an unwanted growth in the size of the battery, and the frequent charge and discharge operations adversely affect battery life. To circumvent the challenges mentioned above, researchers have suggested the development of combined energy storage solutions that merge the capabilities of both batteries and supercapacitors. The supercapacitor aims to increase the range that electric vehicles can travel, improve their dynamic performance, prolong the lifespan of batteries, and mitigate the strain on batteries during rapid energy spikes by exploiting the high instantaneous power capability of the supercapacitor. Energy management strategies are formulated based on the capabilities of the hybrid energy storage system comprising batteries and supercapacitors, aiming to allocate the optimal power output from the battery and supercapacitor for improved vehicle performance, energy efficiency, and battery lifecycle. There are several energy management strategies in battery and supercapacitor hybrid electric vehicles, which include Dynamic Programming, Equivalent Consumption Minimization Strategy, and Model Predictive Control. The state-of-the-art method is the iii reinforcement learning (RL) based energy management strategy. This includes Q-learning, Double Q-learning, Deep Q-networks, and Deep deterministic policy gradient methods, which have been studied in the problem of energy management for the electric vehicles equipped with hybrid energy storage system. However, there are still challenges in RL-based electric vehicle energy management strategies that need further study. First, the RL methods need many iterations, from 15000 to 150000, to converge. The RL methods mimic human brain activity to use experiences to update the agent, which causes a long training time and computational burden. Thus, reducing the number of iterations becomes a critical challenge for RL-based energy management strategies. Furthermore, there is a lack of study on the real-time implementation of RL algorithms for energy management using existing hardware from vehicles. Although the real-time hardware implementation of conventional energy management strategies is abundant, the existing hardware in vehicles is challenging to meet the high-performance computing requirement of RL-based energy management strategies. Advanced hardware products, like GPU from NVIDIA and TPU from Google, are always used in computer vision, natural language processing, and AI supercomputing, and it is not economically viable to apply these expensive products for electric vehicle energy management. Developing the relevant feasible performance learning techniques is very important to reduce hardware implementation costs for electric vehicle energy management. This dissertation addresses these challenges and seeks to contribute to the field in several ways. Firstly, a dedicated driving cycle for EVs is developed, providing a realistic iv representation of driving conditions. Based on this EV driving cycle, an energy management strategy is developed by Q-learning to increase energy efficiency and minimize battery aging. Then, an advanced energy management strategy is designed by imitation learning to decrease the learning time and computational cost associated with Qlearning. Furthermore, a novel Lithium-Sulfur battery with bilateral solid electrolyte interphase is studied and adopted to lower the operating cost of EVs. Lastly, to solve the continuous control problem, a deep reinforcement learning-based energy management strategy is introduced, which incorporates the digital twin technology for real-time implementation. Through these contributions, this dissertation seeks to contribute to the comprehension and practical implementation of energy management methods within the hybrid energy storage systems utilized in electric vehicles. The research findings hold the potential to drive more sustainable and efficient electric vehicle technology while considering practical implementation and cost-effectiveness
Recommended Citation
Ye, Yiming, "Reinforcement Learning-Based Energy Management for Electric Vehicle Application" (2024). All Dissertations. 3567.
https://open.clemson.edu/all_dissertations/3567
Author ORCID Identifier
0000-0001-7737-6100