Research On Probability Prediction Of Ultra-short-term Wind Power Considering The Correlation Of Meteorological Factors

With the continuous upgrading of smart grid and the full penetration of renewable energy,large-scale wind power grid connection has faced more severe challenges.Realizing accurate and reliable wind power prediction plays a very important role in power grid security,economic dispatching and wind power consumption.At present,the point prediction accuracy of wind power power in China is to be further improved,and the information given by the single predictive result cannot quantify the uncertainty of wind power.Therefore,based on improving the accuracy of point prediction,the probability prediction of wind power is further studied to improve the reliability of probability prediction.Aiming at the problems that the historical information is not fully utilized and the multivariable input can not reflect the different feature importance in the current wind power point prediction research,the feature attention mechanism is introduced into one-dimensional convolutional neural network short-term and long-term memory network.An ultra short-term wind power prediction method based on multivariable data is proposed,which can dynamically consider the correlation between different meteorological factors and wind power.In the current wind power point prediction process,the historical information is not fully used and the multidimensional input weight is fixed,ignoring the difference of the importance of different features.Based on the random forest algorithm,the importance of meteorological factors such as wind speed,air pressure,humidity at different heights on wind power output power,combined with accumulation contribution rate screening meteorological features to reduce input dimension.Taking the denoised data of singular spectrum analysis as the input,a cascade multivariable prediction model is established to predict the ultra short-term wind power.The example analysis results show that the method can adapt to the dynamically adjustment of mining characteristic relationship dynamically adjust the weight of each input characteristics,and strengthen the focus on the key characteristics of the prediction time,high point prediction accuracy.In order to quantify the uncertainty of wind power with high accuracy,a hybrid probability prediction method is proposed.The method integrates quantile regression,attention mechanism convolution neural network-short-term and long-term memory network.Based on the different quantile prediction results obtained by quantile regression,attention mechanism,convolution neural network short-term and long-term memory network,the interval prediction with 90%confidence level and 80% confidence level is constructed,and the probability density at 90%confidence level and 80% confidence level is estimated by gaussian kernel density.The experimental results show that the probability prediction method proposed can obtain narrower average width of prediction interval,higher interval prediction accuracy and more reliable probability prediction results than the quantile regression of gated cycle unit and quantile regression of long-term and short-term memory network.