Date of Award
8-2016
Document Type
Dissertation
Degree Name
Doctor of Philosophy (PhD)
Department
Automotive Engineering
Committee Member
Dr. Zoran Filipi, Committee Chair
Committee Member
Dr. Simona Onori
Committee Member
Dr. Pierluigi Pisu
Committee Member
Dr. Robert Prucka
Abstract
This dissertation integrates battery thermal management and aging into the supervisory control optimization for a heavy-duty series hybrid electric vehicle (HEV). The framework for multi-objective optimization relies on novel implementation of the Dynamic Programing algorithm, and predictive models of critical phenomena. Electrochemistry based battery aging model is integrated into the framework to assesses the battery aging rate by considering instantaneous lithium ion (Li+) surface concentration rather than average concentration. This creates a large state-action space. Therefore, the computational effort required to solve a Deterministic or Stochastic Dynamic Programming becomes prohibitively intense, and a neuro-dynamic programming approach is proposed to remove the ‘curse of dimensionality’ in classical dynamic programming.
First, unified simulation framework is developed for in-depth studies of series HEV system. The integration of a refrigerant system model enables prediction of energy use for cooling the battery pack. Side reaction, electrolyte decomposition, is considered as the main aging mechanism of LiFePO4/Graphite battery, and an electrochemical model is integrated to predict side reaction rate and the resulting fading of capacity and power. An approximate analytical solution is used to solve the partial difference equations (PDEs) for Li+ diffusion. Comparing with finite difference method, it largely reduces the number of states with only a slight penalty on prediction accuracy. This improves computational efficiency, and enables inclusion of the electrochemistry based aging model in the power management optimization framework.
Next, a stochastic dynamic programming (SDP) approach is applied to the optimization of supervisory control. Auxiliary cooling power is included in addition to vehicle propulsion. Two objectives, fuel economy and battery life, are optimized by weighted sum method. To reduce the computation load, a simplified battery aging model coupled with equivalent circuit model is used in SDP optimization; Li+ diffusion dynamics are disregarded, and surface concentration is represented by the average concentration. This reduces the system state number to four with two control inputs. A real-time implementable strategy is generated and embedded into the supervisory controller. The result shows that SDP strategy can improve fuel economy and battery life simultaneously, comparing with Thermostatic SOC strategy. Further, the tradeoff between fuel consumption and active Li+ loss is studied under different battery temperature.
Finally, the accuracy of battery aging model for optimization is improved by adding Li+ diffusion dynamics. This increases the number of states and brings challenges to classical dynamic programming algorithms. Hence, a neuro-dynamic programming (NDP) approach is proposed for the problem with large state-action space. It combines the idea of functional approximation and temporal difference learning with dynamic programming; in that case the computation load increases linearly with the number of parameters in the approximate function, rather than exponentially with state space. The result shows that ability of NDP to solve the complex control optimization problem reliably and efficiently. The battery-aging conscientious strategy generated by NDP optimization framework further improves battery life by 3.8% without penalty on fuel economy, compared to SDP strategy. Improvements of battery life compared to the heuristic strategy are much larger, on the order of 65%. This leads to progressively larger fuel economy gains over time.
Recommended Citation
Zhang, Xueyu, "Supervisory Control Optimization for a Series Hybrid Electric Vehicle with Consideration of Battery Thermal Management and Aging" (2016). All Dissertations. 2293.
https://open.clemson.edu/all_dissertations/2293