Research On Seasonal Photovoltaic Power Plant Data Restoration Based On Improved Clustering

The monitoring system of photovoltaic power station needs to monitotr the state of photavoltaic power generation at all times in order to analyze and process the feedback data correctly.However.in actual production,there are often cases of false or missing monitoring data,so it is necessary to study the repair problem of active power output data of photovoltaic power station.In the operation of photovoltaic power station,the size and change of output are affected by geographical location and weather factors,and the influence parameters are difficult to accurately predict.Based on this,this paper studies the method of combining similar power station and similar day.which reflects the influence of geographical location and weather factors respectively to solve the problem of data repair.Firstly,aiming at the analysis of similar power stations,based on the analysis of seasonal variation characteristics of photovoltaic output,this paper proposes an improved clustering method based on effectiveness index to extract seasonal similar power stations.Through the seasonal clustering of actual photovoltaic power station output data.the experiment shows that the improved k-means clustering of Kalinsky halabasz index has the best effect,The clustering speed is the fastest,and the total number of effect indicators is higher than,that of the unimproved clustering.Then,in view of the problem of similar day extraction,this paper uses the existing "similar day" calculation method for reference,and adopts the comprehensive similarity method combining weighted Euclidean distance and weighted grey correlation degree to study the historical similar day extraction of repair day.This method takes into account the number value and change trend of weather feature vector,Similar days were extracted for most of the repair days,and the comprehensive similarity was higher than the threshold of 0.95.Finally,aiming at the problem of repairing photovoltaic defective data,based on the analysis of the time series characteristics of photovoltaic output data,this paper uses the gating recurrent unit neural network model to repair the bad data of photovoltaic power station with reference to the output data of similar power stations and similar days,and calculates the error rate between the repair data and the actual monitoring data,The error rate of the repair results is less than 7%,and the repair effect is ideal.