Abstract: This paper proposes a prediction method that integrates Gaussian process regression (GPR) model. First, the signal is analyzed in time and frequency domain, and the signal time-frequency domain is extracted. Then, we combined with distance correlation coefficient (distance correlation, DC) to select features; in order to improve the prediction accuracy of the model, Bagging algorithm is used to integrate multiple base learners GPR to minimize the impact of noise signals; The weight of each sub-model is obtained by the calculation of the posterior probability, and the output of the sub-model is fused to obtain the global prediction value. A comparative analysis experiment is performed to verify the effectiveness of the method. Compared with artificial neural networks and support vector machines, the method has better prediction accuracy and has certain engineering practical significance.