Application Of Support Vector Machine (SVM) In Short-term Photovoltaic Power Prediction

As the global energy shortage and environment pollution problem increasingly prominent, solar photovoltaic power generation has become a emerging industry which the nations of the world pay attention to and develop because it is clean, safe, convenient, efficient, etc.. But with the intermittency and randomness which solar photovoltaic power generation has,when it is incorporated into the power grid,it must create problems for the safe and stable operation of power system and the power quality,etc.These problems will limit the photovoltaic development.Photovoltaic power prediction is the effective way to solve the problem. Based on the above,through the analysis of the correlation data of photovoltaic power,this paper uses the method based on support vector machines(SVM) to predict short-term photovoltaic power.Support vector machines(SVM) is a new type of machine learning,which can achieve the smallest actual risk by structural risk minimization criterion.lt can well solve the small sample problems,nonlinear problems,high dimensional problems and the problems which may has a local minimum point,etc.It has a strong generalization ability. Based on support vector machines(SVM),this paper utilize method 1 in which the phase space reconstruction and support vector machine (SVM) are adopted,method 2 in which wavelet transform and support vector machine (SVM) are adopted,and method 3 in which Multi-source information fusion thought is adopted to predict short-term photovoltaic power. In method l,this paper deals with the power series through the phase space reconstruction first, then it does the prediction by SVM. In method 2,this paper deals with the power series through the wavelet transform first, then it does the prediction for each branch by SVM,in the final,we can get the prediction by the superposition. In method 3,based on data fusion thoughts,considering all the information about short-term photovoltaic power into the algorithm,this paper deals with meteorological data,power series of this unit own and power series of the adjacent unit by SVM,then it get the final prediction by the neural network. Using field data to do the simulation and comparing the prediction effect the methods which this paper uses are effective,and it can reduce no adverse impacts on the grid brought by the photovoltaic power into the power grid.