Fault Detection And Classification For Photovoltaic Arrays Based On Data-driven Approaches

As photovoltaic(PV)power generation system usually operates in harsh environment,some faults may occur in PV system.The faults can lead to energy loss and even serious security risks.In order to improve the reliability and efficiency of PV system,effective fault detection approaches are essential.Hence,this thesis proposes fault detection and classification techniques for PV arrays based on data-driven approaches.Firstly,the output data of the PV arrays are processed by data-driven approaches,then the processed data are used to train the fault detection and classification model(FDCM)to detect the typical faults of PV arrays.First of all,a fault dstection and classification approach based on CLUB(CLUstering based on Backbone)algorithm and classification and regression tree(CART)algorithm is proposed.Firstly,a large number of unknown classification data set which collected from PV array are clustered by CLUB algorithm.Then,the clustering results are marked by a few of typical data samples to determine the classification of the data set.Finally,the FDCM is trained by CART algorithm with these data set to realize real-time fault detection and classification of PV array.Secondly,to solve the problem that some faults of PV arrays are difficult to be classified due to the similar data characteristics.A fault detection and classification approach based on linear discriminant analysis(LDA)and support vector machine(SVM)is studied.A variety of characteristic parameters of PV arrays are collected to organize the data set,and the data set are projected by LDA to find the projection direction which make the data set have the best classifiability.Finally,the FDCM is trained by SVM with the projected data set.Finally,the characters of principal component analysis(PCA)and LDA are researched,and a hybrid discriminant analysis(HDA)approach is obtained.HDA can further improve the classifiability of data set.And the processed data is used to train the FDCM by SVM.The simulation and experiment results show that the data set projected by HDA can be used to train the FDCM whose classification accuracy is higher than the data set projected by LDA.The CLUB,LDA,PCA and HDA algorithms which belong to the multivariable statistical analysis algorithm of the data-driven approach can extract the features of data set more effectively.And the CART and SVM which belong to the artificial intelligence algorithm of the data-driven approach are used to train the FDCM to realize the fault detection and classification for PV arrays.The researches described in this thesis are verified by the data collected from a simulation PV array and a PV power generation platform with the peak power of 1.8k W.The verification results prove the effectiveness of the approaches.The work of this thesis can provide some ideas for the development of fault detection and classification for PV arrays.