The SOC of Lithium-ion battery prediction based on transfer learning

Transfer learning can solve the problem of low modeling accuracy due to the lack of training data. The SOC (State of Charge) estimation of lithium-ion power battery based on transfer learning can solve the problem that the data of vehicles are difficult to collect under some working conditions. It also can solve the problem of SOC estimation method in practical application and reduce the operation cost of enterprises. This paper designs a lithium-ion battery SOC prediction model based on transfer learning strategy. A comprehensive feature evaluation index combining DTW (Dynamic Time Warping) and JS divergence (Jensen - Shannon divergence) is proposed to obtain the transferable features between source and target domains. The domain invariant features between the source domain and the target domain are obtained by the mixed scale measurement method, and the combined domain adaptation technology is proposed for the adaptation problem of the distribution difference between the source domain and the target domain. A multi-level correction method is adopted to match the difference between source domain and target domain. Finally, the improved transfer learning model is compared with the non-transfer-learning model. The results show that the improved transfer model has higher accuracy and stronger robustness.