Accurate wind power prediction is an important method to guarantee the power system containing large-scale wind power to be safe and stable. This paper analyses the multi-step rolling prediction mode, establishes the ANFIS (adaptive neuro-fuzzy inference system) prediction model, and realizes real-time multi-step rolling prediction of wind power. Taking the real-measured data from a wind farm in the west of Jilin province as an example, case study is done. The subtraction clustering algorithm is used when it form initial fuzzy inference system structure. This algorithm is effective to avoid the combination explosion problem of the artificial setting structure method. The real-time multi-step rolling results of wind power based on the linear regression method, the moving average method and the persistence method are compared with the prediction results based on the proposed ANFIS prediction method. The result of case study shows that the prediction accuracy of the latter method is the highest, and it further illustrates the validity of the ANFIS model.