Short-term Wind Speed Prediction And Extreme Value Estimation Based On NWTC2 Meteorological Data

The demand for renewable and clean energy is expanding day by day,and the development of wind energy as a new type of energy is also growing rapidly.Wind energy is widely used in electric power system,heating system and so on,so the prediction of wind speed and estimation of extreme value distribution have become the primary problem to be solved by using wind energy.The improvement of wind speed prediction accuracy plays a positive role in reducing the cost of wind power generation.Wind speed extreme value distribution fitting estimation is also widely used in wind resistant design of buildings in strong wind areas,and the research on extreme value estimation of wind speed data is bound to increase.In this paper,the correlation analysis of variable data in NWTC2 wind field is firstly carried out,and then the chaotic characteristics of the wind speed of meteorological data are determined.All data were de-noised by DBSCAN clustering,and the most suitable wind speed data sequence was selected for empirical study.According to the GRACH model,a prediction model of EGARCH time series with 1 asymmetric item and a first-order lag item was established.Under the fact that its data are nonlinear,the nonlinear self-regression network with exogenous input(NARX)model is established to make short-term hourly prediction of wind speed data.In addition,in view of the chaotic nature of wind speed data sequence,c-c method is adopted to determine the required parameters in the phase space reconstruction,and chaos support vector machine model is established in combination with chaos theory to predict the wind speed value of the next 24 hours.Finally,the prediction effect was evaluated according to the RMSE and MAPE accuracy of each prediction model.The results show that the least square support vector machine model based on chaos theory has the best prediction effect on the wind speed of NWTC2 meteorological station.In this paper,the extended Burr XII distribution,Gumble distribution and Weibull distribution are used to simulate the extreme value distribution of wind speed data.Firstly,the average excess function of MEF was selected to determine the threshold value of wind speed data.Due to the meteorological characteristics of excessive left part of wind speed data,local data larger than 3m/s were selected to construct the MEF function,then the threshold value was determined.According to the Bayesian-Monte Carlo parameter estimation method,different parameters in the three distributions are estimated.Finally,distribution fitting is carried out for the data.According to the SSE and precision as the index of goodness of fit,the fitting effects of the three distributions are compared and analyzed.The results showed that Gumble distribution had the best fitting effect,but this experiment proposed to introduce the extended distribution of Burr ? into the study of wind speed extreme value,which provided more possibilities for the future research of meteorological data.