Research On Structural Fault Diagnosis And Class Classification Of Photovoltaic Power Station Based On Random Forest

Photovoltaic power stations are prone to failures,which affect their power generation efficiency.The intelligent photovoltaic power plant fault diagnosis model can effectively detect faults,but when the diagnosis model is established,because the fault data is less than the normal data,it will cause the problem of data imbalance during algorithm training,which will affect the accuracy of model diagnosis degree.This thesis focuses on the problem of data imbalance in fault diagnosis.The following three aspects are included in the main research content package:1.The causes and performance of photovoltaic power plant failures are studied,and appropriate failure simulation schemes are designed.In the simulation software,the mathematical model of the photovoltaic power station under different conditions is established.The simulation model simulates the photovoltaic power plant structure under normal operation and the power plant structure under various fault conditions.Their output characteristic curves are obtained and compared and studied,so that the characteristic parameters of the fault diagnosis model are determined and used for training fault diagnosis.The data set required by the model is acquired.2.Aiming at the problem of unbalanced fault data in photovoltaic power plants,a photovoltaic power plant fault diagnosis method based on the combination of AFCM(Alter-Native Fuzzy C-Means)-SMOTE(Synthetic Minority Over-Sampling Technique)algorithm and random forest algorithm is proposed.The AFCM-SMOTE algorithm processes the fault samples,generates "artificial" samples,and uses the "artificial" samples to train the random forest algorithm,and finally the diagnosis of photovoltaic power plant faults is realized.3.In the simulation software,four types of faults:open circuit,short circuit,aging and shadow are simulated.The simulated data is cleaned to form a data set.The AFCM-SMOTE-RF algorithm is used to learn the data to construct a photovoltaic power plant structure diagnosis model.The models established by the AFCM-SMOTE-RF algorithm are compared with random forest,neural network,and SRF(statistical random forest)respectively.Comparative experiments on the simulated photovoltaic power station structure and various photovoltaic power station diagnostic algorithms show that:(1)The photovoltaic power station structure established in this paper is slightly different from the actual power station,with an average analogy degree of 95.2%,which can be applied to photovoltaic power station fault analysis research And data set acquisition;(2)The detection accuracy of the model constructed by the AFCM-SMOTE-RF algorithm is 96.5%.When the AFCM-SMOTE-RF algorithm is applied to photovoltaic power plant fault detection,it solves the problem of inaccurate classification of random forest in photovoltaic fault detection applications because of the lack of fault sample data,and improves the accuracy of fault diagnosis.