A novel hybrid data-driven method for li-ion battery internal temperature estimation

Accurate battery internal temperature estimation is a key for safe battery operation of electric vehicles. In this paper, a novel hybrid data-driven method combining a linear neural network (NN) model and an extended Kalman filter (EKF) is developed to estimate the internal temperature of a LiFePo4 battery. In order to select the proper input terms of the linear NN model and estimate the associated parameters, a fast recursive algorithm (FRA) is firstly used. Then an EKF with a battery lumped thermal model as the state function is used to filter out the outliers and reduce the errors in estimating the internal temperature based on the linear NN model. The test results from two different experiment data demonstrate that the hybrid method can achieve good estimation accuracy, and the method can be easily applied to other type of batteries.