A new data-driven diagnostic technique for in-situ capacity prediction of lithium-ion batteries

Lithium-ion battery (LIB) charging behavior is of equal importance as discharging with abundant information for the battery management system to determine the state of charge (SOC), state of health (SOH), and state of safety (SOS). However, to the best of the authors' knowledge, little work can be found regarding the fast and accurate prediction of the charging behaviors of LIBs. To fill in this gap, in this work, a data-driven diagnostic technique, long short-term memory in situ capacity prediction (LSTM-ICP), is proposed to predict battery capacity through short-term voltage measurements under constant current charging. Unlike the technique in the previous work, LSTM-ICP does not rely on the interpretation of voltage-time data as incremental capacity (IC) or differential voltage (DV) curves. Thus, the need for differentiating voltage-time data (amplifying the process of measuring noise) and the voltage measurement range including the peak of the IC/DV curve can be ignored and thus significantly shorten the prediction process. The optimization method of the mini-batch gradient descent may reduce the computational complexity and also enhance the randomness of prediction. Results highlight the performance of the proposed diagnostic technique, LSTM-ICP is applied to two datasets, which are composed of four and eight cells. In each case, the evaluation results of different performance indicators in a certain voltage range show that the prediction model has good accuracy and effectiveness.