In order to improve the accuracy of short-term wind power forecasting and quantify the uncertainty of power prediction, a combination forecasting method based on Gaussian Process Regression (GPR) and Bootstrap Aggregation (Bagging) is proposed. For the instability and large computing of GPR, Bagging and Fully Independent Training Conditional Approximation (FITC) are introduced. Meanwhile, a new weight strategy based on Bayesian Committee Machine (BCM) is raised. Experiments show that, GPR method which based on Bagging and FITC is better than the traditional method in the stability, precision and training time consuming. Furthermore the improved GPR can get a more accurate confidence interval, and the prediction accuracy of the proposed method also has improved significantly compared with ELM and LSSVM in the wind power prediction.