Abstract: In order to better monitor the health status of lithium power batteries, this paper proposes a method for predicting the health status of lithium batteries based on an improved bidirectional temporal convolutional network, a long short-term memory network, and an attentional mechanism, and uses the crested porcupine optimizer to find the optimal hyperparameters of the method. The proposed method is tested on the University of Maryland lithium battery charge/discharge dataset, and the capacity-related health features are extracted, and the health features with high correlation are screened by the Pearson correlation coefficient and used as inputs to the method. The experimental results show that the method can achieve high accuracy in the health state prediction of lithium batteries. Specifically, the root-mean-square error of the method in the health state prediction of 4 batteries does not exceed 0.020, the average absolute error does not exceed 0.017, and the coefficient of determination is greater than 0.995.