Study On Methods And Evaluation Of Short-term Power Combination Forecasting Of Wind Farm

There are many research productions about wind power prediction.These results show that different prediction methods can be used to reflect different characteristics of the original data.Only combining these methods together,it is able to comprehensively and reasonably use the information to build a model with high prediction quality.In this paper,based on the common short-term power forecasting methods and the analysis of historical data,the short-term power forecasting method for wind farm is studied and evaluated.The main work is as follows:(1).Through the analysis of wind speed and power history data,it is found that the inertia,randomness and the numerical value of wind speed in the wind field,and then analyzes the factors that affect the wind speed.By drawing the power time curve,it is known that the power can follow the wind speed well,so the conversion model of wind speed and power can be established.But in addition to wind speed,temperature,wind direction and other factors also affect the power of the wind,so,these factors can be added to the ranks of the feature vector when the model established.(2).The theory foundation and principle of support vector machine are introduced,and the regression model of support vector machine is deduced;Based on the least square support vector machine(LS-SVM)model,the wind speed of the wind farm in the province of Inner Mongolia is predicted.This paper firstly analyzes the factors that affect the output of wind speed,and then handle the input vector.The particle swarm optimization algorithm is used to find the optimal parameter ? and ? in the LS-SVM model to improve the accuracy of prediction.The running results show that the wind speed forecasting model has good convergence,high accuracy and fast training speed.(3).In this paper,the forecasting model of wind speed-power curve by actual datas of wind turbine generator and wind speed-power curve based on radial basis function neural network are established.After that,the article analyzes the application of the two models and evaluate the operating results with mean absolute error(MAE)and mean absolute percentage error(MAPE).The results show that each model has its own advantages,only by combine these methods together,the quelity of forecast can be better.(4).In this paper,two kinds of prediction methods are combined by entropy weight method,and the results are analyzed and evaluated by MAE,MAPE and absolute error.The operation results show that the error of patttern assembly is smaller than that of single model.Further more,if we shorten the interval of data samplingtime and Re prediction with the patttern assembly,the prediction effect of the model will be better and more suitable for the wind farm.In order to prove the universality of the model,in this paper,multiple sets of data are tested and checkout and these tests have shown good results.So,the prediction model is suitable and can be used for the local wind farm.