Short-Term Wind Power Forecasting Based On Wevelet Transform And Support Vector Machine

In recent years, wind energy has obtained the country’s strong support and encouragementas a clean and sustainable energy, wind power has also developed rapidly. However, powergrid infrastructure is relatively weak in China, when the proportion of its power generation in thegrid reaches a certain level,it would bring austere challenges to the stability of the grid. In orderto solve the shock to the grid wind power generation system as the random speed wavecharacteristics, it is need to accurately predict for short-term wind speed.This paper design a combination of comprehensive prediction model based on wavelettransform and support vector machine theory. First, it uses wavelet transform to obtaincharacteristics of wind speed sequence at different scales, and extracts low frequencycomponent and high frequency wave component in wind speed. Secondly, it uses improvedparticle swarm optimization to optimize parameters of support vector machine. And then, theinput low frequency component of wind is mapped to a high dimensional space by supportvector machine optimized, which realizes the accurate prediction of the wind through waveletreconstruction. Finally, the wind sequence predicted is substituted in characteristic power curveof the fan, then the final prediction result of the wind power will be got.The test show that the model has some advantages such as a great generalization ability, ahigh forecast precision and so on.700training samples,200test samples, the wind sequencepredicted root mean square error lowest can be reached0.0932. The result of predictive powerwhose deviation is0can account for48.5%of the total sample and with the forecast lengthincreases, the error gradually enlarge. It has obviously stronger prediction effect than traditionalprediction model. Compare the two support vector machine prediction model that whether or notgo through parameter optimization, it can prove improved particle swarm optimizationsubstantially improve the prediction accuracy of the support vector machine indeed, and realizeshort-term wind power forecasting of1-3days.