With the extensive promotion of new energy vehicles, the state assessment of their core componentpower batteries and the accurate prediction of rechargeable capacity (RC) are of considerable significance for evaluating the reliability, driving range and residual value of new energy vehicles. This paper presents a prediction method for the rechargeable capacity of new energy vehicle power batteries based on the ITPA-Informer model. Firstly, the rechargeable capacity is estimated by the ampere-hour integration method in combination with the Kalman filter, and two-stage feature engineering (recursive feature elimination and kernel principal component analysis) is employed to select features and reduce dimensions to alleviate the curse of dimensionality in actual working conditions. regarding model, an improved time pattern attention (ITPA) mechanism is introduced in the decoder of the Informer model to extract features at different time scales apart from the sampling time interval. The contribution of each time step to the current prediction is adjusted by an exponential decay factor to enhance the temporal dependency of the rechargeable capacity gradually decreasing with the increase of driving mileage. The experimental results indicate that the proposed model outperforms traditional CNN, LSTM and GRU models in multiple evaluation metrics, and the operation data in different months verify that the model possesses good generalization ability.