Probabilistic Prediction Analysis And Research On The Wind Power

In recent years,the wind power generation industry has been developing at a high speed in our country and around the world.However,due to the volatility and randomness of the wind power itself,the total installed capacity of wind power has brought a series of problems referring to the safety and stability of power grid.If the wind power can be accurately and effectively predicted,it can reduce the capacity of the system to balance the power fluctuations,thereby reducing the cost of power grids and increasing the level of wind power consumption.Because of the limited precision of wind power point prediction and high uncertainty,this paper mainly makes a probabilistic prediction and further study of wind power.First of all,this paper briefly introduces the status quo of new energy power generation industry,pointing out the importance of wind power forecasting and the practical engineering value as while as presenting in detail the achievement of wind power forecasting at home and abroad and its existing problems to be improved,then the time-dependent analysis of the wind power output characteristics is made and the probabilistic intervals forecasting of wind power based on Moving Block Percentile Bootstrap(MBPB)method is proposed.This method avoids the model error introduced by the parameter selection in the traditional parameter model.At the same time,the time-dependent characteristics of the wind power data are taken into account to improve the prediction accuracy.Finally,based on the operating data of a wind farm in the northwestern China,the simulation results show that it has better engineering value by comparing with other methods.In addition,in order to reduce the dependence on the existing point prediction model and optimize the output of the power direct prediction model,a method of interval estimation of wind power probability based on ensemble empirical mode decomposition(EEMD)and kernel extreme learning machine(KELM)is proposed.This method decomposes the wind farm data into sub-sequences with different fluctuation frequencies and amplitudes at different scales by EEMD,and establishes sub-models of KELM power prediction for each sub-series.For the sake of achieving higher prediction performance,particle swarm optimization(PSO)is used to optimize the output weights of the KELM training model,and the prediction results of each sub-model are added to obtain the final power prediction interval.Compared with PSO-EMD-KELM model,PSO-KELM model and PSO-ELM model respectively,the results show that the model has obvious advantages in prediction accuracy and prediction accuracy.Finally,in view of the fact that the traditional prediction intervals estimation index can not reflect the matching of the fluctuation characteristics of the predicted value and the actual value on the time scale,a double-layer wind power interval prediction model is constructed.In the inner model,the comprehensive optimization function covers three evaluation indexes of power root mean square error,power gradient and power curvature so that the prediction converges in the direction of decreasing error and matching fluctuation characteristics.Particle swarm optimization algorithm is used to optimize the output weights of KELM model to get single point prediction power.The outer model takes the difference between the predicted power output by the KELM model and the actual power to obtain the power prediction error sequence.Relying on this sequence,the prediction error probability density function based on the non-parametric kernel density estimation is constructed.Under the given confidence level,the upper and lower sub-quantiles of the probability density function of prediction error are obtained,and then the final predicted power interval is achieved according to the actual power.Since the non-parametric estimation method reduces the subjectivity of the model selection,it can better grasp the dynamic structure of the sample sequence and improve the accuracy of prediction result.