Study Of Short-term Wind Power Prediction

The capacity of China’s wind farm has been greatly improved compared with the past inrecent years. However, large-scale wind power has brought serious challenges to the safeoperation of the power system due to the volatility of wind power. Wind power prediction isone of the key technologies to solve this problem through predicting the generating windpower of the next period of time to support the scheduling and control of the power grid andreduce the impact of the wind turbine output power fluctuations on the power grid.At present, most researches use the statistical methods to establish the power predictionmodel. This article uses artificial neural networks and support vector machine to researchwind power prediction modeling, and the research finds that BP,RBF, GRNN of artificialneural networks and linear kernel, polynomial kernel, RBF kernel of support vector machinecould not predict well when the actual power fluctuates rapidly, which limits the wind powerprediction accuracy. To solve this problem, this article applies a modeling ideas: First,decompose the actual power sequence into orthogonal components with different fluctuationscales; Secondly, establish prediction model of single separate component; Finally, weighteach single component prediction model output to get the final predict power. The main workof this article is:1) Comparing BP, RBF, GRNN of artificial neural networks and linear kernel, polynomialkernel, RBF kernel of support vector machine, polynomial kernel of support vector machineis selected for prediction modeling;2) Using EMD to decompose wind power time series, and selecting SVM to train everysingle component of wind power time series as single component prediction model;3) Introducing IOWA to the field of wind power prediction, and proposing an improvedIOWA weighting strategies for wind power prediction.Based on the modeling ideas, within test set prediction accuracy of improved IOWAweighting strategies performs better than other modeling methods.Based on the research results above, selecting SVM to train prediction models of somewind power prediction system and it is applied in some wind farms across the country.