Study On Data Characteristics Extraction Of Wind Farm And Wind Power Real-time Forecast

Wind power has the characteristics of high randomness and volatility, whichmay cause voltage or frequency deviation, voltage fluctuation and even taking offthe power grid after the large scale wind power connecting to power grid. Accuratewind power forecast can provide safe, economic and high-quality operation forpower system. From previous research works, it can be obtained that the wind speeddistribution, the importance characteristics of affecting factors and the similaritycharacteristics of training sample have important influences on forecastingperformance. Through extraction and analysis of those three data characteristics ofsome typical wind farms, this work is exploring the correlation between those threedata characteristics and prediction accuracy, and provide theoretical basis for theselecting reasonable and effective modeling method, optimize the forecasting modeland improve the wind power prediction accuracy further. For that, this paper hasdone the following works:1) For the problem that different prediction ways produce different forecastingaccuracy and computation, based on the wind speed distribution characteristics inwind farm, a novel approach for wind power real-time forecasting is proposed. Thesimulation results for real data of wind farm at two typical areas in China, coastalareas and North China, show that the prediction accuracy is improved and theprediction computation is significantly reduced with the presented method.2) On the analysis of the importance characteristics of affecting factors and thesample similarity characteristics, an integrated neural networks approach combiningfuzzy rough set and improved clustering is proposed in this paper. Firstly,collecting wind farm information and building wind speed knowledge system, fuzzyrough set was applied to carry on the attribute reduction for a variety of factors affecting wind speed to optimize the model input, and the importance of eachattribute for wind speed was obtained. Then, the traditional clustering wasimproved through the weighted Euclidean distance based on attributes’ importance,and similar data were extracted as the training set to optimize the training set.Lastly, matching model was selected to carry out the wind speed prediction. Takinga wind farm in north China as an example, the results show that the method caneffectively improve the forecasting accuracy with a small model input.