Clustering LiFePO4 cells for battery pack based on neural network in EVs

Power sources of electric vehicles (EVs) are integrated by hundreds of cells connected in series and parallel. Due to slight differences in the manufacturing process and various operating conditions of the individual cell in a battery pack in EVs, the characteristics of the individual cell to cell can be non-uniform in the battery pack, which affects the available capacity and the life of the battery pack. This paper investigates the neural network (NN) based method to sort lithium iron phosphate (LiFePO4) cells so that the cells can be as consistant as possible before these cells form a battery pack. All cells used in the sorting process are LiFePO4 Cells. A clustering NN tool of self-organizing map (SOM) is adopted to classify the cells based on the experimental data of all cells. The cell temperature and capacity are used in the trained SOM to classify the cells. The cells in the same category are connected in series to form a sorted battery pack whereas the randomly chosen cells are connected in series to form the unsorted battery pack. The comparison between the sorted battery pack and the unsorted battery pack has been conducted to verify the clustering methods and the consistency of the cells in the battery pack.