Research On Short-term Wind Power Interval Prediction Algorithm Based On Similar Sample Optimization Selection And Its System Implementatio

The protection of the ecological environment has received more and more attention in the international community,and China has also proposed the corresponding "3060" dual-carbon goal to build a low-carbon economy.As an indispensable part of building a low-carbon economy,the development of green renewable energy has great strategic significance.As a renewable energy industry,wind power is developing rapidly around the world.With the increasing scale of wind farms connected to the grid,the uncertainty of wind power still exists,which is still a potential huge challenge to the whole power system.Wind power system is a system with intermittency,randomness and volatility.The main reason is that the uncertainty of natural factors has a great impact on it.Therefore,accurate and detailed wind power forecast data has become an increasingly important factor in the operation of wind farms and power grids.At the same time,the traditional point prediction cannot track the actual power curve and meet the requirements of some power peaks.In practical applications,reasonable interval prediction can help power system dispatchers to make reasonable decisions and dispatching.As a result,the short-term wind power prediction method is deeply studied from the perspective of sample number and interval prediction,and the main work is as follows:Based on the principle of "similar day" in power load forecasting,a short-term wind power forecasting method with multi factor similar samples is proposed to improve the accuracy of wind power forecasting.The main characteristic factors are selected by kernel principal component analysis(KPCA)method,and the weight is determined.Then the similar days are matched by dynamic time bending distance method(DTW),and the overall similarity is defined by multi-element weighting.The training samples of the model select the appropriate sample data of similar days according to the overall similarity,and construct the short-term wind power prediction model combined with long-term and short-term memory network(LSTM).The experimental comparison shows that,it is not difficult to find that the above has greatly improved the prediction accuracy and model efficiency of short-term wind power.Secondly,on the basis of improving the deterministic prediction accuracy of wind power,in order to characterize the uncertainty in wind power,an interval prediction of short-term wind power is carried out.The nonparametric kernel density estimation method is bounded by pseudo-data method.At the same time,in order to improve the local adaptability of the kernel density estimation method,an adaptive window width optimized by the gray wolf algorithm is introduced.Finally,the probability density curve is converted into a power density curve,and the upper and lower limits of the power are obtained.The experimental results clearly draw the following conclusions.The improved kernel density estimation method has high accuracy,interval coverage and narrow interval width.Finally,the above algorithms are integrated in the actual wind power prediction system,and displayed in a visual way to provide reference for system users.