Review on Prediction Methods of Remaining Useful Life of Lithium Ion Batteries

Abstract

Compared with other types of ion batteries, lithium-ion batteries have great advantages in specific capacity, self-discharge rate, performance and price. Therefore, lithium-ion batteries have developed rapidly and are widely used in aerospace, portable equipment and so on.The rul prediction method of lithium-ion battery is developing from experimental reliability to practical reliability, expanding the application scope of the prediction method and ensuring reliable ex ante maintenance, which can not only ensure the safety of personnel, property and equipment in civil use. There are two main methods for rul prediction of lithium-ion battery: one is to build a capacity decline model based on the internal electrochemical mechanism of the battery, Due to the complex internal chemical mechanism of the battery, the differences between the same batch of batteries lead to its application limitations; The second is the data-driven method. Through the extraction of parameters in the process of battery operation, such as voltage and current, support vector machine, artificial neural network and other methods are used for prediction. From the prediction results, the data-driven method has high precision and wide application. It is the mainstream research method at present.It is of great significance to accurately determine the health status of lithium-ion batteries. To address the problem that the prediction of a single limit learning machine algorithm is prone to jumping, the method of using artificial fish swarm optimization to optimize the limit learning machine is proposed to try its best to predict the model of the remaining life of lithium-ion batteries. Firstly, the isovoltage discharge time is extracted as an indirect health factor, then the limit learning machine is optimised using the artificial fish swarm algorithm to build an indirect prediction model for the remaining life of Li-ion batteries, and finally a validation evaluation is carried out based on the NASA dataset B0005-B0006. The experimental results show that the proposed model predicts stable prediction results with high accuracy and small error in prediction results.