The improvement of wind speed prediction accuracy plays a positive role in reducing the cost of wind power generation and arranging the location of wind farms. The DBSCAN clustering method is used to denoise all data and select the most appropriate sequences for empirical research. Firstly, since the chaotic wind speed data sequences would affect the prediction results, C-C method is made to determine the needed parameters in phase space reconstruction. Meanwhile, a model of chaos support vector machine combined chaos theory is established to predict the wind speed value in the coming 24h. Then the model is compared with the EGARCH model and the nonlinear self-regression network with exogenous input (NARX) model in terms of prediction results. Finally, the prediction effect of models is evaluated by the RMSE and MAPE of each prediction model. The results show that the support vector machine model based on the chaotic time series has the best prediction effect on the wind speed of the NWTC m2 weather station. This work is supported by National Natural Science Foundation of China (No.11731012 and No.11571052) and Natural Science Foundation of Hunan Province (No.2018JJ2417).