Lithium battery state of health prediction method based on online knee point warning identification of coupling temperature

The capacity of lithium-ion batteries degrades rapidly after the knee point, which means that early warning before the knee point is more important for the continued safe use of the battery. This paper proposes a dynamic early identification knee point warning framework. Pearson correlation analysis and mutual information are used to prove the dependence between temperature and incremental capacity. Copula-Based Outlier Detection is used to evaluate the coupling effect. Moving T test is used to determine the knee point warning. Secondly, the similar knee point warning batteries under similar charging and discharging conditions are searched in the cloud database, so as to train the Gaussian process regression prediction model. In this paper, a total of 30 batteries are used. The knee point warning appears 15.12% earlier than the knee point on average, and the average root mean square error of the state of health prediction result is as low as 0.76%.