State of health estimation of lithium-ion batteries based on multi-health features extraction and improved long short-term memory neural network

Abstract

Accurate state of health estimation of lithium-ion batteries is essential to enhance the reliability and safety of a battery system. However, the estimation accuracy based on a data-driven model is degraded by one health feature and incorrect hyper-parameters selection. This paper develops a battery state of health estimation method based on multi-health features extraction and an improved long short-term memory neural network. To accurately describe the aging mechanism of batteries, health features are extracted from battery data, such as time features, energy features, and incremental capacity features. The correlation between multi-health features and state of health is evaluated by the grey relational analysis. Aiming at the problem that the hyper-parameters of an neural network model are difficult to select, an improved quantum particle swarm optimization algorithm is developed to correctly obtain the hyper-parameters. The experimental results show that the mean absolute error, mean absolute percentage error, and root mean square error of this method are all within 1%, which is much lower than other methods, with high state of health estimation accuracy and robustness.