Study On Variance Prediction Of Wind Power Random Fluctuations

Wind speed prediction is an important technical means to ensure large-scale wind power integration. The time resolution of the wind speed forecast is 15 min and the forecast results are average wind speed in a sense. But the actual wind speed is composed of the average wind speed and the wind speed instantaneous random fluctuation. On the basis of the average wind speed forecasts,the study of wind speed instantaneous random fluctuation can provide more detailed information about real-time wind speed, which is contribute to schedule a more detailed control strategy to stabilize wind power fluctuations. In addition, the wind speed instantaneous random fluctuation is also an important factor of the wind turbine selection and safety design. Therefore, this article focus on wind speed instantaneous random fluctuation, the main work is as follows:First, we define the concept of variance of wind speed instantaneous random fluctuation to study it. Meanwhile, the calculation method of wind speed variance is proposed based on Wavelet Analysis. The real-time monitoring data of wind farm is used to calculate wind speed variance. Then the physical characteristics of wind speed variance is studied. The results demonstrate that: there is a multiscale modulation effect between average wind speed and wind speed variance; the diurnal variation curves of wind speed variance exist distinct diurnal cycle phenomenon.Power system need to master the wind speed information ahead of time in order to stabilize wind power fluctuations. Therefore, it is significant to predict the wind speed variance. The premise of wind speed variance forecast is that it has predictability. Therefore, the predictability analysis method of wind speed variance is put forward based on the correlation analysis theory of time series analysis. The analysis results show that the predictability length of wind speed variance is approximately between 1h and 5h, which indicated it can be forecast. The crosscorrelation between wind speed variance and average wind speed is studied as well.Thirdly, the best input dimension of prediction model is selected by experiments. Then the forecast of 10 min ahead, 20 min ahead, 30 min ahead, 40 min ahead, 50 min ahead and 60 min ahead of wind speed variance are conducted, respectively. And the forecast results are evaluated by MAE and MSE. The stability and generalizability of the forecast models are confirmed by the wind speed variance data of Heilongjiang wind farms in 2013. In order to improve the forecasting performance of the original model, the information of average wind speed is added to input. Experiments show that this method can improve the accuracy of prediction results in certain conditions.Finally, prediction errors of the models is analyzed. The forecast errors increased along with the increase of the forecast step length. Therefore, the selection of prediction step length is critical in the prediction, otherwise the reliability of the forecast results would reduce. T Location-Scale Distribution, Normal Distribution and Extreme Value Distribution are used to fit the distribution of forecast errors. Based on a large number of statistics, it’s found that t location-scale distribution is suitable to identify the probability distribution of forecast errors. At the same time, the distribution of the forecast errors is proved to be independent of the forecast model, whether it is established based on BP neural network or support vector regression(SVR).