Battery pack consistency modeling based on generative adversarial networks

In working condition of battery packs, the battery pack consistency has a great impact on the overall performance of the battery pack. In order to build an accurate battery pack model, we need to build a battery pack consistency model. Firstly, we used a Gaussian mixture model to fit the statistical characteristics of a single parameter. This method can accurately fit the skewness in the parameter distribution and fit the multi-peak characteristics that may appear. Secondly, we constructed a nonparametric battery pack consistency model using a Generative Adversarial Networks (GAN). Our consistency model can accurately describe the statistical characteristics of a single parameter and fits the correlation coefficient between parameters. The battery pack model substituted into the GAN-generated battery parameters exhibits a very high similarity to the experimental data. The relative errors of the simulation results are less than 0.6 % for the terminal voltage and less than 0.3 % for the energy utilization efficiency (EUE), proving the advantages of the GAN consistency model in fitting the distribution of the battery parameters. Finally, we implemented the GAN consistency model in an embedded system with limited computing resources, which proves that our proposed model has the ability to run normally on existing BMS. (C) 2021 Elsevier Ltd. All rights reserved.