The Autoregressive Moving Average Model (ARMA) predicts the wind speed around the telescope to lessen the impact of random and time-varying wind disturbances. This gives the telescope control system enough movement time to compensate for structural deformation. Combined wind speed prediction and servo control systems help reduce the pointing jitter caused by wind disturbance when the antenna performs observation tasks under wind disturbance and improve the pointing accuracy. This research analyzes the meteorological data, wind speed, and direction collected by the wind measurement tower at the Xinjiang 110 m aperture radio telescope (QTT) site to obtain the seasonal wind speed variation. The wind direction of the QTT site is primarily concentrated in the south and north. For different seasons, the wind direction characteristics are not apparent in summer and autumn and are primarily concentrated in the south-southeast direction in winter. For non-stationary wind speed time series, the proposed model uses the Augmented Dickey-Fuller Test (ADF) and Kwiatkowski-Phillips-Schmidt-Shin Test (KPSS) to verify the stationarity of wind speed time series in the training set or observe whether the autocorrelation coefficient decays quickly to zero with the lagged value. Based on the data characteristics of the wind field at the QTT site, the Akaike information criterion (AIC) and the Bayesian information criterion (BIC) are used to identify the model order. Through the maximum likelihood estimation to estimate the model parameters, the model tests the validity of the autoregressive moving average prediction model of wind speed in different quarters at the QTT site by analyzing whether the residual characteristics of the model meet the normal distribution. Three error indexes are selected to examine the accuracy of prediction model: root means square error (RMSE), mean absolute error (MAE), and means absolute percent error (MAPE). For the MAE of the model prediction data and the test data, the MAE of the summer prediction model is 0.133mcdots^-1, the MAE of the autumn prediction model is 0.162mcdots^-1, and the MAE of the winter prediction model is 0.287mcdots^-1. These results of the prediction error of the ARMA established based on the wind speed data for different quarters of the QTT site verify the accuracy and stability, which proves that the ARMA can achieve high prediction accuracy and reliable effects. The model meets the needs of wind disturbance control of the radio telescope and provides a systematic scheme for wind speed prediction in practical and necessary data support for the wind disturbance control of the radio telescope.