According to the problem of weight determination of every model in combined wind power prediction, a probability-based weight and optimization combined method is presented. Firstly, different prediction models are optimized and combined based on the feature of homogeneity between weights and probability. Then the optimal weight under different wind field for every model is determined. At last, the combined model is improved to dynamical prediction model for higher prediction accuracy. Experimental results indicate the proposed method based on probability can effectively improve the accuracy of short-term wind power prediction. The combined prediction model has certain reference value for wind power prediction.