State of health estimation for lithium-ion batteries based on fragmented charging data and improved gated recurrent unit neural network

When employing data-driven methods to estimate state of health of lithium-ion batteries, the aging feature extraction and a large amount of training data preparation are often indispensable. To mitigate this concern, an improved gated recurrent unit neural network model is developed to achieve accurate state of health estimation of lithium-ion batteries with easier feature extraction and limited training data. Firstly, the charging duration of two certain voltage increase intervals is extracted as health features, and the grey relational analysis and Spearman's correlation coefficient are employed to evaluate their relevance and rationality, respectively. Subsequently, a gated recurrent unit neural network model is constructed, of which the key hyperparameters are adjusted and optimized based on the whale optimization algorithm. Finally, the effectiveness of the proposed model is demonstrated by comparing its estimation results with different algorithms. The results indicate that the optimized gated recurrent unit neural network model constructed can accurately estimate battery state of health with the truncated training data, and leads to the average error of less than 1 %. Moreover, the model is validated on different types of batteries, demonstrating preferable generalization capability.