Research On Ultra-Short-Term Wind Power Forecasting Model Based On Time Section Fusion

With the rapid social progress and economic development in China,the proportion of wind power generation in the modern power system is getting higher and higher,which will directly affect the quality of electric energy and the stability of the power system operation,and bring great challenges to the regulation and control of the power market and power grid dispatching.At present,accurate prediction of future wind power is an important technical means to solve this problem.Accurate wind power forecasting plays an important role in improvin g the stability and economy of power grid operation and the permeability of wind power.However,due to the fluctuation and uncertainty of wind speed and the geographical dispersion of wind farms,it is difficult to accurately predict wind power at present.This paper mainly studies and improves the ultra-short-term wind power forecasting model.The main work includes several aspects as follows:For data-driven wind power forecasting model,the most important condition to achieve accurate forecasting is to use accurate and reliable original data for training.However,due to wind abandonment,extreme weather conditions,sensor failure and other reasons,a large number of abnormal data exist in the original wind power sequence obtained by us.These abnormal data will seriously affect the reliability of the prediction model,so it is necessary for us to effectively preprocess the wind power data needed for model training.This paper proposes a quarterback based algorithm and wavelet transform algorithm with the combination of wind power anomaly data identification and correction method.Firstly,in view of the abnormal data more scattered,quarterback method is used to distinguish and remove,and then we use the wavelet transform algorithm to identify and correct abnormal data accumulated in wind power curve type.Finally the interpolation method is adopted to reconstruct the missing data,Relatively accurate historical data of wind power are obtained for the research and optimization of power prediction model.The ultra-short-term wind power requires the wind power within the next four hours to be predicted,and the ultra-short-term prediction usually requires a relatively high prediction accuracy.For single performance of wind power prediction model is relatively low,the prediction accuracy can not fully meet the requirements,this paper chose the technology route,wind power forecasting is put forward based on pattern classification and time section super short term wind power combinatio n forecast model of data fusion.First of all,using different machine learning model for wind power parallel prediction.Multiple wind power prediction results under each time section are obtained,and then the fusion mode classification model is used to carry out pattern recognition and classification for the data of each time section.Finally,the data fusion model corresponding to the fusion mode is used to fuse the parallel prediction results into the final wind power prediction results.Considering the limitations of neural network prediction model and support vector machine prediction model,this paper proposes to optimize the parameters of neural network model and support vector machine model based on genetic algorithm to further improve the combined prediction model and improve the prediction effect of the prediction model.Firstly,the basic concept and principle of genetic algorithm are expounded.Secondly,according to the initial connection weight and deviation in the neural network model,it is coded into the genetic algorithm population individual,and then the global optimization adjustment is carried out,and the best parameters obtained are used in the neural network model for wind power prediction.Then the genetic algorithm is used to improve the parameters of SVM model,and the best parameters obtained are used for SVM prediction modeling.Finally,the neural network model improved by the genetic algorithm and the SVM model were used to forecast the wind power in the ultra-short term in the time-section fusion model,and the prediction effect was compared with that of the time-section fusion model without parameter optimization.The simulation example shows that the improved prediction model can improve the learning efficiency of the genetic algorithm and make the prediction effect better.