Prediction And Research On Short-term Wind Speed Of Wind Farm Based On Combination Method

Along with the growth of energy demanding and the increasing lack of fossil fuel, wind power, because it’s clean, circulation renewable, and has many other advantages, becomes one of the most promising ways of power generation. However, wind power possesses characteristics of randomness and intermittency, so the large-scale grid generation increases the difficulty of the power grid schedule. Predicting accurately for the next few hours wind power will be good for electric power dispatching departments to reasonably arrange the dispatching plan of wind power system after the grid generation. Wind speed prediction is the foundation of the prediction of wind farm output power. Therefore, this paper will give further research on hours in advance the prediction method of wind speed on wind farms.This paper first studied wind speed and the distribution characteristics of the wind power. Then, from the basic concept of time series analysis, the paper analyzed the time series prediction trend extrapolation and exponential smoothing method. And from the model and forecasting in detail, it realized three hours ahead an average speed of the forecast. Then, adopting the least squares Support Vector Machine (LS-SVM) as a regression prediction model, it improved standards support vector machine. The nonlinear LS-SVM is to map the data to high dimensional space and establish a linear model to make the calculation speed effectively improved. Finally, based on the two models above, it put forward a short-term wind speed prediction model based on the combination method. This combination model has flexible parameter selection ability, which can make a better use of the wind speed influence information, and can better reflect the internal contact of the wind speed changes.Through comparing and analyzing the wind speed forecast results of a wind farm in Dalian, it showed that the results of combined forecasting model compared with those of one fold model have been improved obviously. Choosing the right combination forecast model and parameter estimation method with the optimum model parameters can effectively increase the prediction accuracy.