A Mechanistic-Data Model based Temperature Estimation Method for Lithium-ion Battery

This paper investigates the problem of lithium-ion battery temperature estimation based on a mechanistic-data model. First, a lumped-parameter thermal model of battery is introduced to provide prior temperature knowledge for the deep neural network. Then, a convolutional neural network (CNN)multi-head self-attention mechanism (MHSA)long short-term memory (LSTM) deep learning model is constructed. By merging the advantages of the lumped-parameter thermal model and the CNN-MHSA-LSTM model, a physics-informed battery temperature model is proposed to achieve high-precision estimation, in which the estimated temperature from the thermal model is used as prior input to refine the predictions of the CNN-MHSA-LSTM model. To enhance the estimation performance of the physical information model, a model hyperparameter optimization method based on osprey optimization algorithm (OOA) is developed. Finally, experiments are conducted in the temperature range of 5C to 25C, and the results demonstrate that the proposed model has a better accuracy improvements compared with the thermal models and data-driven methods, respectively.