With the rapid development of the stock market, the field related to stock prediction has attracted more and more researchers' attention. Stock forecasting can promote the development of stock market well, which is of great significance to the society, enterprises and individual investors. Through experiments, we found that the Support Vector Regression model (SVR) is unstable under the influence of the distribution characteristics of training data, which is specifically reflected by the large deviation in the stock price forecast of some stocks. To solve this problem, this paper improves the SVR model from the perspective of ensemble learning. Based on the SVR model, we integrate two simple and effective models, linear regression model (LR) and K-nearest neighbor model (KNN), to enhance the generalization ability of the SVR model. The experiment shows that the ensemble model proposed in this paper has a significant improvement in the accuracy of stock price prediction compared with the simple SVR model.