Multi-feature fusion based battery health estimation using bidirectional long short-term memory networks

Battery state of health (SOH) estimation has always been a challenge in designing battery management systems. Due to the safety and long cycle life, Li-ion batteries have been widely used as energy storage carriers. In this paper, we propose a multi-feature fusion-based battery SOH estimation based on bidirectional long short-term memory neural networks. According to the aging mechanism of the battery, the IC curve is extracted and filtered, and then the peak value is extracted. Extract energy and capacity from a fixed voltage range of charging data. These three features are utilized as health features to train and test the neural network model. In addition, the correlation between these three features and SOH was analyzed to justify the feature selection. The feasibility of the proposed method was verified by estimating the battery module capacity and calculating the evaluation metrics. In addition, the superiority of the method is demonstrated by comparing with other models (e.g., back propagation neural network, multiple linear regression).