State of Charge Estimation by Joint Approach With Model-Based and Data-Driven Algorithm for Lithium-Ion Battery

Abstract

In order to ensure the safety and service life of lithium-ion batteries for automotive applications, accurate state of charge (SOC) is required for system management in the process of driving. It is particularly challenging to estimate the SOC by using the online approach, for which the battery dynamics model is nonlinear. Many researchers have focused on model-based or data-driven algorithms alone, but comparatively few of them use a joint approach with the two types of algorithms. The data-driven algorithm is self-learning and has better adaptability, while the model-based algorithm is more stable and has stronger robustness. If these advantages can be combined, a better SOC estimation approach will be developed. In this article, based on battery charge dynamics, a complex fractional-order model of battery is simplified into a discrete fraction model for engineering application of control algorithm. A Bayesian belief network (BBN) is used to estimate the battery model parameters, and the adaptive extended Kalman particle filter (aEKPF) is used to estimate the SOC. In order to obtain accurate parameters of battery model for training, linear programming is used to identify the parameters online. Collectively, this article designs a joint approach of how the aEKPF with BBN estimates the SOC precisely. A developed approach has been downloaded into a battery control unit and tested in real-world conditions using a battery test bench to realize practical applications of the joint approach.