Study Of Short-term Wind Power Prediction Based On Least Squares Support Vector Machines

The fluctuation and instability of wind power output, caused by the fluctuation andIntermittent of wind power, has a great influence on wind power integration and powergrid scheduling, which contributes to serious impact on large-scale wind power accesssystem. It is very important to make accurate predictions of output of wind farm power,for the purpose of reserving wind power to the utmost extent, under the premise ofsafeguarding security and stability operation.In this paper, the Least Squares Support Vector machine is chosen to build ashort-term wind power prediction model. In terms of the great influence on theperformance of the prediction model the kernel function and the related parameters ofleast squares support vector machine have, this paper establishes least squares supportvector machine model, based on different kernel functions for wind power prediction, andselects an optimal one from all the kernel functions, and uses Gravitational SearchAlgorithm to optimize the parameters of the model. The prediction results are comparedbetween support vector machine model and BP neural network prediction, which showsthat the least squares support vector machine model based on ERBF kernel function andGSA has better prediction accuracy. It indicates that this model is a more excellent windpower prediction model.For conventional prediction, only deterministic point prediction can be made, whilethere is a level of prediction error for wind power. In order to determine the probability ofoccurrence of a predicted value, the wind power is predicted, using uncertain probabilisticforecasting, and then the confidence interval for each prediction point forecasting can bebuilt, therefore, a sufficient support for electricity scheduling and risk analysis can beachieved. In this paper, we use non-parametric estimation methods to predict the windpower in a short-range, calculate the variance and deviation of prediction points based onthe linear smooth property of LSSVM regression model and finally we build theconfidence interval of short-term wind power forecasting under different confidence level.Based on the single-point and interval prediction method for short-term wind powerprediction, we simulate the model before-mentioned, Matlab, and design a Matlabgraphical user interface for wind power forecasting system in order o facilitatevisualization operation of the forecasting algorithm. This system is based on Matlabinterface design platform GUIDE and can realize the single-point prediction and interval prediction of wind power. Furthermore, a simple and beautiful user interface and menu baris designed which can realize the modules such as data input, model selection. In the end,the forecasting results and prediction error is displayed by graph.