Remaining useful life prediction of lithium-ion batteries based on phase space reconstruction and optimized LSTM network

To solve the problems of insufficient adaptability and low prediction accuracy for lithium-ion battery remaining useful life (RUL) prediction, this paper proposes an optimized neural network prediction method to improve prediction ability. Firstly, the original signal is used to decompose and reduce noise by Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN), and extract the residual that best reflects the degradation trend as a feature. Secondly, the phase space reconstruction method can determine the most appropriate sliding window size and reconstruct a new sample space to improve prediction accuracy. Thirdly, the genetic algorithm is applied to perform an adaptive global search for the optimal key parameters of Long Short-Term Memory (LSTM) network, which improves the adaptability of the method on RUL prediction. In order to verify effectiveness, the model was actually applied to NASA batteries, and the experimental results show that the proposed method has higher accuracy and generality than other traditional methods.