The state of health (SOH) of lithium-ion batteries (LIBs) is a critical parameter of the battery management system. Because of the complex internal electrochemical properties of LIBs and uncertain external working environment, it is difficult to achieve an accurate SOH determination. In this study, we proposed a novel SOH estimation method by using prior knowledge-based neural network (PKNN) and Markov chain for a single lithium-ion battery. Firstly, we extract multiple features to capture the battery aging process. Due to its effective fitting ability for complex nonlinear problems, the neural network with prior knowledge-based optimization strategy is adopted for the battery SOH prediction. The Markov chain, with the advantageous prediction performance for the long-term system, is established to modify the PKNN estimation results based on the prediction error. Experimental results show that the maximum estimation error of the state of health is reduced to less than 1.7% by adopting the proposed method. By comparing with the Group Method of Data Handling (GMDH) and back propagation neural network in conjunction with Levenberg-Marquardt algorithm (LMNN), the proposed estimation method obtains the highest SOH accuracy.

State-of-health (SOH) of lithium-ion batteries plays a vital role in the safe and reliable operation of electric vehicles. However, most of the existing SOH estimation methods still require a large number of battery aging data while the established model usually lacks generalization. Here, we build a data-driven transfer learning model to obtain more generality on the SOH estimation. First, potential health features are extracted from battery charging data and then pruned via the importance function. Second, support vector regression (SVR) is employed to establish source models with different battery data, which takes selected features as the input and capacity as the output. Third, the transfer-stacking (TS) method is utilized to combine all source models. A TS-SVR method for SOH estimation is then established only using the first 30 $\%$ of target battery data after solving the optimization problem of assigning weight to each source model. Finally, the proposed algorithm is verified by three different battery datasets and shows better estimation performance than the comparative algorithms. It is proved that the proposed method uses only a small amount of target battery data while together with the source battery data can achieve an accurate SOH estimation during its life cycle.