Objective: Lithium-ion batteries have been widely used in electric cars as the source of power with its advantages of high energy density. The development and use of electric vehicles are affected by the performance degradation of the battery with cycling and ageing. However, the present researches on the battery cycle life are mainly focused on the single cell rather than battery packs, which is different from the actual situation of electric vehicles provided energy by battery packs. This paper examines the degradation characteristics of the battery pack during the charge and discharge cycle and suggest effective model to characterize the capacity degradation and predict the remaining useful life (RUL) of lithium-ion battery pack.
Methods: An experiment was carried out for researching the degradation characteristics of the ternary lithium battery pack which is used in a electric car. The battery pack is tested for 400 charge and discharge cycles, in which the voltage and capacity data of the battery pack are collected through battery management system and test equipment. Random vector functional link (RVFL) network model is trained to track the pack's degradation trend over its cycle life based on experimental data analysis. Compared with RVFL, several approaches such as statistical models, Kalman filtering and Particle filtering are also applied to predict the remaining useful life of the pack.
Results: From the analysis of experimental data, the remaining useful life of the battery pack drops from 37.271 Ah to 36.094 Ah after 400 cycles. At the end of discharge, the voltage distribution of the batteries in the battery pack is inconsistent, with a maximum difference up to 0.28V. When individual batteries reach the upper limit of charge or lower limit of discharge more quickly, the energy of the battery pack cannot be fully utilized and the remaining useful life is limited. The prediction results show that, Kalman filter and particle filter have errors in battery RUL prediction due to the form of state function. It can be seen in the prediction results at cycle 300 that RVFL model is more accurate to predict the battery pack remaining useful life trend.
Conclusion: Different from the remaining life of a single battery, the inconsistency of battery capacity within a battery pack is an important factor limiting the overall battery pack life. The degree of capacity inconsistency of the batteries in the battery pack will further increase with cycling and ageing of the pack. Data-driven prediction models, such as SVM, neural network models, when performing predictions in 100 cycles, the amount of data available for training is insufficient to accurately predict RUL. Model-driven Kalman filtering and particle filtering prediction methods are limited by the form of model equations. As the number of cycles increases, Kalman filtering and particle filtering prediction methods are less accurate than SVM and RVFL.
Mr. Weijian Li, South China University of Technology, CHINA; Prof. Dr. Fengchong Lan, South China University of Technology, CHINA; Prof. Dr. Jiqing Chen, South China University of Technology, CHINA; Dr. Yigang Li, South China University of Technology, CHINA