Remaining useful life prediction for lithium-ion batteries based on a hybrid model combining the long short-term memory and Elman neural networks

This paper presents a novel hybrid Elman-LSTM method for battery remaining useful life prediction by combining the empirical model decomposition algorithm and long short-term memory and Elman neural networks. The empirical model decomposition algorithm is employed to decompose the recorded battery capacity verse cycle number data into several sub-layers. The recurrent long short-term memory and Elman neural networks are then established to predict high- and low-frequency sub-layers, respectively. Comprehensive battery test datasets have been collected and used for model parameterization and performance evaluation. The comparison results indicate that the proposed hybrid Elman-LSTM model yields superior performance relative to the other counterparts and can predict the battery remaining useful life with high accuracy. The relative prediction errors are 3.3% and 3.21% based on two unseen datasets, respectively.