Research On Nonparametric Kernel Density Estimation Method Of Wind Power Forecast Error

With the introduction of measures to accelerate the development and utilization of clean energy during the 14 th Five-Year Plan period,the scale of grid-connected wind power across the country will inevitably continue to expand.At the same time,the randomness and volatility of wind power generation has brought more serious safety hazards to the power system.It is possible to accurately grasp the probability distribution of wind power forecast errors to ensure power safety.Existing research results show that the error distribution characteristics of wind farms in different regions are quite different,and the selection of parameter models depends on the subjective judgment of the engineers,which encounters greater difficulties in practical applications.Compared with parameter estimation,non-parametric estimation does not require prior model assumptions and is a more applicable method in theory.At present,unbiased cross-validation(UCV)and rule of thumb(ROT)are two commonly used nonparametric kernel density methods,but due to the peak and thick tail in the wind power prediction error,and for local small sample features,the direct use of these two methods will produce larger generalization errors and reduce the accuracy of the non-parametric kernel density model.Therefore,exploring more accurate optimization methods for error probability distribution has practical significance for the application of nonparametric estimation in wind power prediction error modeling.According to the current research status of wind power prediction error modeling,the paper has done the following research: First,study the application range of UCV and ROT in wind power prediction error modeling,and use random samples to simulate their peak thickness Under the condition of tail error distribution,the accuracy changes with the number of samples and shape parameters.Second,in view of the problem of increased generalization error caused by insufficient sample completeness and normality,a wind-based method based on the smooth bootstrap method is proposed.Power prediction error kernel density modeling method,given the bandwidth optimization model and implementation process that introduces weight coefficients,and uses the multiple bisection test method to verify the practicability of the method;third,the non-parametric estimation method causes the tail accuracy and skewness to appear relatively for the problem of large errors,the paper uses the advantages of the generalized Pareto distribution in reflecting the probability distribution of extreme events,and proposes an optimization model of semi-parametric density estimation.The simulation of simulated data and real data shows that the semi-parametric estimation model proposed in the thesis is correct and practical.