Research Of Hybrid Model Of Multilayer Perceptron And Support Vector Machine In The Wind Power Prediction

Renewable energy has attracted the wide attention all over the world,with the increasingly prominent problems of energy crisis and environmental pollution.Wind,a clean and green renewable energy,has been widely used for power generation.Statistically the proportion of wind power generation in Chinese overall power system,which is increasing year by year.Due to the growing maturity of wind power generation technology,wind power installed capacity and the scale of grid connected gradually.However,due to the influence of weather,environment and other factors,wind power shows volatility,randomness and intermittency.The large-scale connection of wind power will affect the stability,safety and reliable operation of the power grid seriously.Therefore,develop an accurate model to predict the power of wind farm will be conducive to the power system scheduling,the power balance and normal operation.As communication failure,system control or manual intervention,some history data will show abnormal changes when the wind power data is collected from the wind power station,which will have serious impact on analyzing the statistic characteristics of wind power.In order to reduce the negative interference of abnormal data on the fluctuation and the prediction of wind power,Firstly,according to pyramid data classification model which is presented by the theory of the geometrical topology and data standardization,not only can we pretreat the abnormal data,but also can we classify data,which means we can assign consistent labels to the same kind of data for further prediction modeling.Secondly,due to the power of the wind power is a set of time sequence showing the characteristics of a certain period,in fact,using the data with the same or similar characteristics as the model input and output of training set is more beneficial to the improvement of the prediction precision.So in this paper,by comparing the six feature selection methods for the model of training,We select the most suitable feature selection method for the training set of the model to select input and output data with the same or similar characteristics.Finally,a novel hybrid prediction model named self-iterative partially connected model(MLP-SVM)is proposed based on multi-layer perceptron(MLP)and support vector machine(SVM).Due to the defect of the multi-layer perceptron(MLP)on the hidden layer transfer equation of the input and output,SVM needed to be introduced in order to optimize MLP.Then,the transmission equation determined by the fully connected backward propagation algorithm of the general neural network and the proposed highway law determines a mixed prediction MLP-SVM model.To test the validity of the model,this research collected the wind power data of 5 northwest cities(Jiuquan,Mazongshan,Minqin,Wuwei,and Zhangye)from July 2014 to March 2015 to examine.The result indicates that the proposed model is featured with better predictive accuracy and validity compared with the classic model.