Techniques for Battery Management: Modeling, Estimation, Learning & Controls
- Park, Saehong
- Advisor(s): Moura, Scott J.
Abstract
This dissertation develops an advanced battery management system where lithium-ion battery dynamics are governed by electrochemical principles, and estimation & control techniques are applied to enhance the performance and longevity. The rapid progress in electrified transportation, electronic devices, and energy storage system needs cutting-edge management system that can monitor and control the electric battery system safely. We address this topic in three phases: i) modeling, ii) estimation, and iii) controls respectively.
First, we provide an overview of battery fundamentals and develop mathematical models for electrochemical battery phenomena. Electrochemical modeling is an important tool for simulating the performance of a battery. With the electrochemical model as a simulator, one can demonstrate prototypes of battery management system without conducting expensive experimental protocols. Furthermore, designing reduced-order electrochemical models helps to develop real-time estimation/control algorithms for large-scale battery systems.
Second, we propose a systematic framework to identify battery cell using the electrochemical model. To accurately represent the cell chemistry, the model parameters must be correctly identified. Good fitting parameters take a long time to identify. Often, it requires disassembling the cell and performing specialized electrochemical characterization tests, a costly and difficult process. The proposed parameter estimation framework for an electrochemical model enables to estimate electrochemical parameters by using input-output data.
Lastly, we then develop a novel battery charging strategy for the identified electrochemical model while accounting for the electrochemical constraints associated to aging mechanism. The model-based optimal solutions reveal that the characteristic of optimal charging protocol is Bang-Ride control, which provides theoretical supports on the existing charging strategy. Furthermore, learning-based controller design via reinforcement learning is highlighted as an adaptive controller for battery charging strategy. The performance of this learning-based feedback controller is demonstrated with other charging strategies.