STATISTICAL MODELS AND INFERENCE FOR LI-ION BATTERY PROGNOSTICS

Abstract

Developing prognostics and health management methods for Li-ion batteries has received increasing attention in recent years. This thesis proposes three statistical models and inference for the Li-ion battery prognostics based on the easy to measure operational profiles. The three models include a Bayesian hierarchical model which is good at long term predictions of battery degradation state, a state space based model which is appropriate for short term predictions, and a hybrid model which combines a physical and statistical model. With the developed models, we can take full use of battery operation profiles, implement battery in-cycle operation management and remaining useful life prediction in one framework and update prognostic results with real time observation. The effectiveness and promising features are demonstrated by practical case studies.