| The proportion of photovoltaic power generation worldwide is increasin g year by year.With the continuous development of technology,the database and data volume of power plants are increasing.So data analysis is imperative in the optim al operation of po.wer plants.Electric energy itself has the particularity that it can only be supplied in real time.When the data quality of power station is poor,it will interfere with the next data application,such as power prediction,generation planning,etc.And a large number of accurate real-time data plays a vital role in the safe operation of power grid and real-time scheduling.Therefore,timely solution of data defects is of great practical value to ensure the security and reliability of power system operation.Firstly,the regularity of output po,wer fluctuating with temperature,weather characteristics and illumination intensity is analyzed.The types of bad data and the causes of bad data are listed,and the identification method based on Leyte criterion for abnormal fluctuation data is determined.Then,considering the spatial correlation of power data of different power stations,and using similar power stations can make up for th e data defects of the power stations to be repaired.This paper introduces the basic principle and steps of hierarchical clustering,and makes a simple analysis of the shortcomings of traditional clustering in similarity screening.Finally,the similarity recognition method considering the degree of distance correlation and the degree of curve shape correlation is determined,It improves the accuracy of clustering extraction of similar power stations,and lays a foundation for the establishment of follow-up data repair model.Secondly,the grey correlation method is used to analyze the meteorological information,and the repair model is established with similar power stations.The principle of selecting training samples and the principle of building the whole repair model are analyzed.The structure of neural network model for different bad data types is given,which provides a basis for simulation examples.Finally,the simulation analysis of Qinghai power stations is carried out.The output curves of the modified power station and similar power station are listed respectively,which fully shows that the similarity between the similar power station and the modified power station is high.According to the types of bad data in power plants,the all-day data loss,local data loss and abnormal fluctuation data are repaired,and the root mean square error and relative error are analyzed.The results of the restoration fully show that the accuracy of the repair results has been greatly improved compared with the traditional similar day method after considering the auxiliary correction of similar power stations. |
