Lithium-Ion Batteries SOH Estimation With Multimodal Multilinear Feature Fusion

Abstract

As a key status for battery energy storage systems, state of health (SOH) can provide the fundamental information for lifespan management of the battery pack in electric vehicles (EV). Traditionally, features are extracted from various signals to establish a powerful data-driven model for the lithium-ion battery (Li-ion) SOH estimation. One issue left is how to utilize the existing features from diverse modalities of measurement signals for a superior battery aging information capture. The available options are selecting the features by analyzing their correlations with SOH. This paper aims to investigate the intercorrelations between various features through the multimodal multilinear fusion mechanism, which enables to utilize the multimodal multilinear features (MMF) and their interaction characteristics. A high-order polynomial module is designed to fuse the MMF from various sources. To improve the efficiency and performance of the SOH estimator, a 2D convolutional neural network (CNN) network is chosen to use the proposed MMF. The performance of the proposed method is validated on two independent datasets, which obtains the lowest mean absolute error (MAE) of 0.37% and the lowest root mean square error (RMSE) is 0.45%.