Study On Short-term Wind Speed Forecasting Of Wind Farms

Wind power, as a clean and renewable energy, has received more attention and been developed strongly all over the world. However, due to the randomness of wind speed, the wind power output is unstable. Large-scale wind power connected to the grid power system has serious adverse effects on the safe and stable ope ration of power system and power quality. Therefore, in order to mitigate the adverse effects, it is extremely necessary to forecast wind power accurately, and wind speed forecasting is the basis of the wind power forecasting. Under such circumstance, this paper launched a profound research on the problem of short-term wind speed forecasting.In this paper, the wind speed forecasting model is established on the basis that the correlation between adjacent time points in the wind time series is high. A support vector machine(SVM) model with similar data is p roposed. In model, similar data which is similar to prediction samples is extracted from lots of historical wind speed data to establish the training samples by a new similar measure function which takes into account the values and trends of wind data. The model is trained by training samples, and finally forcasts the next time wind speed. The proposed modle is used to achieve single-step and multi-step prediction of wind speed respectively and it takes the actual wind data to update the history data in single-step prediction and predictive values in multi-step prediction afrer one step forecasting. The simulation is done in MATLAB, and the results show that the similar data search improves the relevance of prediction samples and training samples, and reduces the prediction error.The paper also studys the wind speed range forecasting. Information granulation is applied to predict the range. To obtain the minimum value sequence, average value sequence and maximum value sequence of the original data changes, h istorical wind data is processed with fuzzy information granulation. The three sequences are forecasted respectively by using the SVM model with similar data. Finally, the range is obtained. The results show that granulated data can reflect the characteristics of wind but also reduce redundant information. By validation with the actual wind speed data of a wind farm, this model can effectively predict the range of wind speed.