| Wind energy has been concerned by researchers and environmentalists because of its green,pollution-free and high reserves.Recently,researchers have become more enthusiastic about developing and utilizing wind energy because it is very consistent with China’s carbon neutralization strategy.However,due to the chaotic nature of Earth’s weather systems,The randomness,fluctuation and load characteristics of wind turbine have an extremely adverse impact on the normal operation of power generation and renewable energy system.It is only by accurately predicting the power of wind that wind farms and power systems can be kept safe.Therefore,based on the actual production data of a wind farm in northeast China,to maximize the economic benefits of wind energy connection,we examine the wind power interval prediction method and the power distribution strategy.The specific research contents are as follows:（1）Production data read from the wind farm’s SCADA system are not only normal operating data,but It is also mixed with abnormal data caused by the discontinuity and uncertainty of wind power.The first step is to preprocess the production data,and then draw the power curve of the actual manufacturing process,which is derived from the wind speed and wind power data,using the power curve,the distribution characteristics and causes of the abnormal data are analyzed.finally,determine the fan under non power limiting operation as the prediction modeling object through the data distribution of the power curve Three sigma laws are used to filter and eliminate the discrete anomaly data in the power curve.In order to ensure the data completion,the missing data is filled in according to the interpolation principle.（2）Data characteristics of historical wind power data are used to select the input variables for the prediction model.Linear function normalization is used for wind speed and wind power data to eliminate the impact of dimensional difference between wind speed and wind power data on prediction accuracy.The Pearson correlation coefficient,random forest algorithm and lasso regression algorithm are integrated to calculate the correlation between the current time wind power data and the past time one,then select the time series characteristics of the wind power input data of the prediction model,decompose the input variables of the prediction model by using the variational modal decomposition algorithm,and the final input variables of the prediction model are determined by data reconstruction.（3）Establish the wind power interval prediction model.The high-precision wind power point prediction results are obtained through the echo state network,and the distribution characteristics of the prediction error are analyzed.The kernel density estimation method with Gaussian kernel function is used to obtain the wind power probability prediction interval with a given confidence of 95%.Several groups of model comparison experiments are designed,and six performance evaluation indexes of MAPE,RMES,MAE,PICP,PINAW,CWC are selected to evaluate the performance of the model in detail.The ESN-KDE interval prediction model can achieve interval coverage of 97% and above.The ESN-KDE interval prediction model can achieve interval coverage of 97% and above.（4）Optimization of power allocation strategy based on wind power interval prediction.The power equalization strategy which does not take into account the generating capacity of each typhoon will greatly increase the wear rate of the fan and reduce the power generation income of the wind field.According to the forecast deviation,a two-objective optimization model is set up in which the total wind field generation is equal to the command generation capacity and the total wind turbine generation capacity are optimized.Differential evolution algorithm solves the optimal weight ratio coefficient and obtains the optimal result of power allocation strategy.Based on the above research,a system applied to the actual production of wind farm is designed. |
PDF Full Text Request