Study on different data-driven models of the lithium battery capacity-fading model.

At present, the lithium-ion power battery system is the core component of new energy vehicles, and an effective battery management system can improve electrical performance and safety performance for the vehicles. It is worth mentioning that battery cycle life is an important indicator of power battery safety. Moreover, the most direct parameter to characterize battery performance degradation is the remaining capacity of a battery. Therefore, remaining capacity prediction of the battery after multiple charge-discharge cycles has become a research hotspot. In a normal working process, the residual capacity of power batteries cannot be measured by direct testing means, so it is necessary to use battery characteristic parameters that are easy to measure to predict the residual capacity. Many scholars at home and abroad extract characteristic values from dynamic charge-discharge curves, but this method is easily affected by battery operating conditions. In order to avoid the influence of dynamic conditions, voltage increase during the resting time of the battery after the discharge period is used as a health factor to characterize the decline of battery capacity. The mapping model between health factors and battery remaining capacity is established by using backward propagation (BP) algorithm and a support vector machine (SVM). The test data of more than 500 cycles of charge-discharge of an in-service 42 Ah ternary lithium battery are taken as samples. Among the samples, the first 400 cycles of the data are set as training samples. Then, through the subsequent 111 cycles of the data, this paper predicts the remaining capacity of the battery by using the model of the BP algorithm and the SVM respectively. The data results and the change trend of the test data in the two predicted models are the same. The MSE for the model of BP and SVM are 1.4% and 0.6% respectively. The experiment results show that, after multiple charge-discharge cycles, there is an obvious linear relationship between the voltage rise sequence during the shelving period and the maximum remaining capacity of the ternary lithium battery. As health factors, voltage rise parameters can be a theoretical reference for the analysis of the remaining capacity of the same type of ternary lithium batteries in practical application. The BP and SVM models can be used to accurately predict the remaining capacity of batteries. Compared with the BP algorithm, the SVM algorithm has better performance when used for small-scale data, with more accurate and effective prediction. [ABSTRACT FROM AUTHOR]