A Data-Driven Battery SOH Estimation Method with CNN-LSTM Model and SSA Optimizing

Abstract

Facing the bottleneck problems of lithium-ion batteries durability, precise estimation of state of health (SOH) is the key of the lithium-ion battery management system. This paper proposes a data driven SOH estimation method on the basis of a convolutional long short-term memory (CNN-LSTM) model. The hybrid model leverages both the advantages of convolutional neural network (CNN) and long short-term memory (LSTM), in which CNN captures special features and reduces the dimension of the charge-discharge data while LSTM preserves temporal information of time sequences. In addition, the sparrow search algorithm (SSA) is integrated to optimize the key parameters including the number of neurons, initial learning rate and L2 regularization coefficients. A simulation experiment including comparisons with the existing classical SOH estimation algorithms is given to show the performances of the discussed SSA-CNN-LSTM method.