Research On Wind Power Prediction Based On Extreme Vector Machine

Large-scale wind power grid brings many challenges to the safe and economic operation of power system because it's strong randomness and intermittent.Wind power prediction technology can provide advance or real-time power output forecasting for power system optimization,and help power dispatching staff to make regular energy output plan,which has great significance to further improve the ability of power system to absorb wind power and is conducive to change China's power structure.However,the wind power prediction accuracy cannot meet the demand of system scheduling,especially when the volatility of wind speed is strong,the traditional forecasting method cannot be used to predict the stochastic scene effectively,which leads to the increase of forecast error.This will bring a significant impact to the power system.Therefore,establish a forecasting model which can deal with the stochastic characteristics of wind power,extract and combine the most effective information among the prediction results is an important means to improve the forecast precision and disperse the forecasting risk.In this paper,three kinds of vector machine method are used to study the wind power prediction,and it is applied to the combined forecasting model.The main work includes:(1)Research on wind power forecasting method based on parallel support vector machineTwo kinds of support vector machine models to wind power predict are established,include: Least Squares Support Vector Machine(LS-SVM)and Parallel Support Vector Machine Model(P-SVM).By analyzing the influencing principle of input quantity to forecast,the model input variables are selected,and the actual operation data of a wind farm in China is taken as an example to verify the model effect.The results show that,the two support vector machines can reach the standard in all seasons,and the P-SVM has better effect to LS-SVM(the error of P-SVM is 14.44% and the error of LS-SVM is 16.08%).The error is largest in spring and autumn which has strong fluctuation,so it is necessary to establish a more predictive model to describe the randomness of wind speed and wind power.(2)Research on wind power forecasting method based on limit vector machineThe wind power prediction model based on limit vector machine is established.The random fluctuation characteristics of the wind power can be described by the learning strategy of the randomly generated hidden layer nodes,whic h increases the simulation accuracy of the stochastic process in the wind power generation process.The results show that the prediction accuracy is better than support vector machine(SVM),and the average RMSE of the whole year is 13.61%.(3)Research on combination forecasting based on three kinds of vector machine algorithmsThe combined forecasting is based on the error characteristics of the three models,use the linear combination method of fixed weight average and nonlinear combination method of multiple attribute decision to the combined prediction respectively,and the model examples have been validated and analyzed.The results show that,the two kinds of combination method is superior to single wind power forecasting model,and the combination of multi-attribute decision making is more effective.