In order to improve the evaluation accuracy of battery health state under the conditions of diversity and small sample data, a diversity-enhanced Stacking integrated learning regression algorithm is proposed based on integrated learning theory. The core idea of the algorithm is to build diversity data through K-means clustering algorithm based on dynamic time warping, and then, Stacking integrated learning regression algorithm is adopted to learn the diversity characteristics of the data, obtain better model accuracy, and enhance the model to diversity data generalization ability. Stacking integrated learning regression algorithm is composed of multiple base learners and an output learner. Firstly, preliminary results are obtained through multiple base learners, and then, the primary results are further studied through the output learners to obtain the final results. Finally, the public battery data set of NASA is utilized to verify the effectiveness of the proposed algorithm.