State of Health Prediction for Lithium-Ion Battery Pack Using a Shared Embedding Layer-Based Quantum Long Short-Term Memory with Transfer Learning

Abstract In this paper, an approach leveraging a shared embedding layer-based quantum long short-term memory (QLSTM) neural network and transfer learning technique is proposed to enhance the accuracy of lithium-ion battery (LIB) pack degradation predictions. The model is especially applied for lithium-ion battery (LIB) aging prognosis when there is limited real-time training data. The framework integrates various methods, including a shared linear embedding layer, QLSTM algorithm, and sliding window techniques. It employs a QLSTM with embedding layers to compress feature dimensions and make predictions. The model is trained offline using historical pack capacity degradation data from a source domain (LIB1) and then transferred and fine-tuned to a target domain (LIB2). Comparative experiments demonstrate the robustness of the proposed model, with the root mean square errors (RMSEs) of pack LIB capacity predictions consistently below 0.002 as the training data increases. The proposed approach exhibits promising results in improving the accuracy of pack capacity degradation predictions for LIB aging prognosis.