The invention relates to a cloud server load prediction method. Aiming at the characteristics of load data, the advantages of the cloud model and the Markov chain are fused to propose a cloud server load prediction method. The method follows the historical data sample training method of the cloud model. In the calculation of the degree of membership value of the predicted value, the predicted degree of membership vector is obtained by Markov forecasting. Then according to the predicted degree of membership vector, the predicted value weighted summarization mode is adopted for the cloud module, so as to achieve the organic combination of the cloud model and the Markov chain. Finally, the load prediction of the cloud server is achieved.