Research On Wind Power Interval Prediction Model Based On Multi-point Numerical Weather Forecast Input

Abundant reserve,clean and environmental-friendly property enables wind power to be intensively developed and utilized.The dispatching and operation of power grid are confronted with an enormous challenge due to the uncertainty of wind power connecting to power grid in a large scale.The prediction of wind power with an increasing accuracy has been regarded as the most direct and effective strategy to cope with the aforementioned challenges.Hence,the present study,describing the Interval prediction model of wind power input by multi-point Numerical Weather Prediction(Numerical Weather Prediction,NWP),is of great practice.Aiming at the influence of abnormal data of wind power on prediction accuracy,this paper preprocesses abnormal wind power data,which are identified combined with the characteristics of different wind power abnormal data.The interpolation method and moving average method are used to reconstruct the abnormal data.Comparing the reconstruction effects with the correlation between data,it is found that moving average method can obtain the best reconstruction effect.Aiming at the influence of model input data on wind power prediction accuracy,the spatio-temporal scale range of wind power prediction model input data is determined on the basis of considering the influence of atmospheric motion process on wind power.The time scale of model input data is determined by the autocorrelation test results of wind power sequence,and the spatial range is determined by the division results of wind farm clusters in the region.For the influence of the calculation ability of the model on the prediction accuracy of wind power,a multi-point NWP input convolutional Neural Networks(Convolutional Neural Networks,CNNs)wind power prediction model are established.Based on the principle of the algorithm and the selection of the spatio-temporal scale range of the input data,the input mode of the input data is introduced,and the training prediction of the prediction model is carried out with the processed output power as the target.Comparing the prediction accuracy of different models,it is found that the wind power prediction model with multi-point NWP input created by CNNs with strong data mining capability has higher prediction accuracy.What's more,a multi-point NWP input wind power interval prediction model is established.The trained CNNs wind power prediction model with multi-point NWP input is used to predict the wind power value in a certain period of time,and the corresponding prediction error is obtained.The distribution of wind power prediction error under different wind power prediction levels is analyzed,and the single parameter estimation model,mixed parameter estimation model and non-parameter joint probability density estimation model of wind power prediction value-prediction error under different wind power prediction levels are established.Combining the above model with the deterministic wind power prediction results,the wind power interval prediction results with a certain confidence level are obtained.The comparison shows that the wind power interval prediction results obtained by the cross-validation method to determine the best window width of wind power prediction value-prediction error nonparametric joint probability density estimation model have the best prediction effect.