State-of-Charge Estimation of Lithium Battery Based on Transfer Learning

While the data-driven methods have been widely applied in State-of-Charge (SOC) estimation, they also expose somes problem of relying on battery data. Depending on different conditions, such as temperature, the battery data will show different characteristics. In order to improve the accuracy of SOC estimation, it is essential to establish data-driven models under different conditions. However, traditional machine learning methods require a large amount of training data. Meanwhile, the cost of obtaining battery data in practice is high. In this paper, transfer learning is applied to SOC estimation modeling to reduce the demand for training data and improve modeling efficiency. Experimental results show that the transfer learning method can not only achieve accurate estimation of battery SOC at different temperatures, but also reduce the amount of training data during modeling. It is also shown that the transfer learning method can realize the rapid modeling of SOC estimation under different conditions when the battery data distribution is similar.