Prognostic modeling of performance degradation in energy storage by lithium-ion batteries

For energy storage, lithium-ion batteries have been widely utilized in cell phones, electric vehicles, and many electrical and mechanical devices. Accordingly, their performance significantly affects these devices' usage experience. This paper presents a particle filter-enabled prognostic modeling method to identify and track battery performance degradation, which exhibits strong nonlinear behavior with time-varying degradation rates. An adaptive resampling strategy has been developed to enable particles to be resampled according to the system's variation. It further solves the particle degeneration problem associated with the standard particle filter, and reduces the number of particles required for system estimation and tracking, thereby improving the computational efficiency. To model battery capacity regeneration that the available capacity of the battery experiences sudden increases, a total variation filter is integrated with the adaptive particle filter to detect such transient performance recoveries, for improved accuracy in performance and remaining life prognosis. To evaluate the developed model, data on battery degradation provided by the NASA Ames Prognostic Center has been analyzed. The result indicates that the model can accurately track battery performance degradation, detect the points in time where performance recoveries have occurred due to capacity regeneration, and predict the battery remaining life, more reliably when compared with other commonly used techniques such as extended Kalman filter and standard particle filter.