Online Lithium Battery Fault Diagnosis based on Least Square Support Vector Machine Optimized by Ant Lion Algorithm

摘要

It is difficult to implement rapid and current fault diagnosis for the lithium battery because of its strong coupling and uncertainty. Moreover, the problem of getting lithium battery fault data will cause a lack of training samples, thus resulting in the unsatisfactory performance of the diagnosis model. Furthermore, once the parameters of the normal fault diagnosis model are confirmed at first, they cannot be readjusted. This will lead to poor generation ability of the model, which is not suitable for the complicated and various lithium battery working conditions. To resolve these above problems, a diagnosis method that is combined with Least Square Support Vector Machine (LSSVM) and optimized by Ant Lion Algorithm (ALO) and Online learning based on sample centre distance, is presented. ALO simulates the process of ant lion capturing ants to implement parameter optimization, which can improve model performance. Online learning based on sample centre distance will remove the samples that have little influence on model training. In this way, ALO will avoid local optimal solutions, and the parameters of the model can be updated according to the new data, thus improving the adaptability of the model to actual working conditions. To verify the possibility and effectiveness of the proposed method, experiments of lithium battery under various fault states are taken. Fault feature data are achieved from these experiments and are used for training, testing and comparing with diagnosis models. The comparative results reveal that the model presented above can guarantee the speed of model parameter updating, showing better generation ability. It's more suitable to implement rapid and current fault diagnosis for the lithium battery.