A hybrid method for online cycle life prediction of lithium-ion batteries

Many industrial applications use lithium-ion batteries, but lack of maintenance, harsh use environments, and poor charging operations accelerate their degradation. Therefore, online remaining useful lifetime (RUL) prediction is a hot research topic. The RUL estimation analysis of a battery can be based on the normalized capacity as the state of health of its cycle life. We propose a hybrid method based on a bidirectional long short-term memory model with an attention mechanism (BiLSTM-AM) model and a support vector regression (SVR) model for online cycle life prediction. Once the sensor collects temperature readings, it uses SVR to update the initial data online to obtain a multistep advance prediction of the temperature and then uses BiLSTM-AM to predict cycle life. The proposed model is verified using 12 lithium-ion phosphate/graphite cells, and the results show that the average RUL estimation error is 3.72%.