Abstract: Power prediction is an important way to address the volatility and randomness brought by new energy generation and enhance the security of power grid dispatching. This paper analyzes the problems brought by new energy generation and the deviation impact of power prediction from different perspectives. This paper also proposes a prediction algorithm(AF-SVM) based on the combination of adaptive filtering and support vector machine(SVM). Herein, adaptive filtering minimizes the mean square error between the output signal and the expected signal by adjusting the filtering weights, thereby optimizing the prediction model. Experiments show that this algorithm performs better in accuracy than traditional power prediction and SVM algorithms.