Transfer learning with CNN-LSTM model for capacity prediction of lithium-ion batteries under small sample
Gang Zou, Zhen Yan, Chengying Zhang, Lei Song
Abstract
Abstract Nowadays, the application of lithium-ion batteries in various industries including our daily life is increasing, so the evaluation of the safety and reliability of lithium-ion batteries is essential. Assessing the performance of lithium-ion batteries accurately and in real-time has become a more critical research direction. However, in most application scenarios, the problem of small sample is often hidden in performance evaluation tasks of lithium-ion batteries. A model-based transfer learning framework to solve this problem is presented in this paper. Firstly, a CNN-LSTM fusion model is proposed by combining the solid spatial feature extraction capability of convolutional neural network (CNN) and the solid temporal dimensional feature extraction capability of long short-term memory network (LSTM); further, a model-based transfer learning approach is introduced to accomplish the performance evaluation task by pre-training and re-training. The experiments are conducted on NASA’s public dataset of lithium-ion batteries, and multiple sets of comparison experiments are set up. The results show that the method can better solve the small sample problem encountered in the lithium-ion batteries performance evaluation task.