A Remaining Useful Life Interval Prediction for Lithium-Ion Batteries Considering Multiple Health Indicators

This paper proposed a new interval prediction method for remaining useful life（RUL） of lithium-ion batteries. From a practical point of view, this method used more easily measured health indicators to realize the interval prediction of lithium-ion battery RUL, and gave a prediction interval with upper and lower limits instead of a single prediction value, which made the prediction results more suitable for realistic decision-making environments. Firstly, multiple groups of health indicators that could characterize battery degradation performance were extracted from the data set. Secondly, the grey relation analysis was used to analyze the correlation between the extracted health indicators and the battery capacity, and two groups of health indicators with the highest correlation were selected for subsequent processing. Then, the fuzzy information granulation was performed on the two groups of health indicators to generate granules, and the upper and lower limit sequences that could be used for interval prediction were obtained. Finally, the upper and lower limits of the granules were treated as the training input and the capacity as the output, the least squares support vector machine was introduced to train and predict lithium-ion battery RUL. The experiment results show that the RUL interval prediction results of the proposed interval prediction method in the three groups of battery data correctly cover the real value, and the prediction intervals are within 30 cycles. Furthermore, in the interval prediction process of four groups of batteries, the coverage values of prediction interval are above 90%, and the prediction width values are within 0.056.