Recognition and Reconstruction of Photovoltaic Output Abnormal Data Based on Geographic Correlation

Abstract

The accurate output prediction curve is based on accurate basic data. Aiming at the problems of abnormal and missing historical data of collected photovoltaic power generation data, this paper takes the light intensity curve and the power curve of adjacent photovoltaic power plants as input references, and proposes a neural network for identifying and reconstructing abnormal photovoltaic power generation data based on geographical location Network model. After analyzing the correlation of each reference quantity, pre-process and normalize the selected light intensity data and neighboring power station data respectively. For missing data, use nnz function to identify. For singular data, a curve similarity function is constructed, and the function values are clustered to be identified by contour coefficients. The sample data point set of [0.5,1] is used as the training set. [-1, -0.5] is the singular point abnormal photovoltaic output curve, plus missing samples, the two constitute the sample data set to be repaired. Finally, the identified abnormal data is repaired by BP neural network. Experimental results show that the method is simple and effective, and the effect is relatively good.