Study On Methods Of Short-term Wind Speed Forecast Based On Support Vector Machine

Energy problem is the major issues to be settled urgently in the process of human sustainable development. With the verge of depletion of conventional energy, renewable energy is receiving increasing attention around the world. Wind energy occupies an important position in the development of renewable energy. Wind power generation has good prospects in renewable energy sources from ripe technology and feasible financial condition.Wind power generation is intermittent and can not be dispatched like conventional power as a result of wind's randomness, fluctuation and uncontrollability. Along with the development of wind power loading, the proportion of wind power in the grid is also increasing. Grid integration of large wind energy will result in many problems, such as power system supply and demand unbalance, the safety of the power systems and the power quality and so on. So, in order to solve those problems, we should provide the prediction of the wind power advance. If wind speed in wind farm can be forecasted accurately, dispatching department can adjust dispatching plan according to the forecasted data ahead of time. So it will lessen the unfavorable influence caused by wind power to the entire electrical network. If the prediction can reach a high level of the precision, it can be beneficial to the power system and improve the competition of the wind energy.Support vector machine (SVM), a new method developed in recent years, is an advanced research field in machine learning. Support vector machine for regression (SVR) has global optimization and well general optimality by approximating non-linear functions with controllable accuracy. This paper tries applying SVR to forecast short-term wind speed.This paper, based on statistical learning theory and support vector machine principle, analyses the characteristics and influences of kinds of support vector regression machine, the kernel function and the parameters of the model. Meanwhile, for the peculiarity of wind speed forecast, such as small samples, nonlinear and high-dimensional regression, this paper brings the methods of support vector regression machine into short-term wind speed forecast, and constructs short-term wind speed forecast model. Additionally, wind speed data in wind farm is taken as an example, which demonstrates detailedly the process of building and solving the model. During the process of selecting prediction model parameters, Grid-search is used firstly for large-scale search to obtain the optimal value( C ,σ), and k-fold Cross validation method is used secondly to choose best parameter combination (ε,v). This paper applies the LIBSVM 2.86 software package to train and predict the training sample data. The analysis of calculation result shows that the model v-SVR has better performance.