Failure mode classification for grid-connected photovoltaic converters

This chapter developed a failure mode classification mechanism for condition monitoring of PV inverters. The developed algorithm performed signal pre-processing by DWT for noise removal, feature extraction and region of interest segmentation. The wavelet coefficients associated with the DWT were optimized by a novel approach based on HSA. Various types of features were extracted once the signal processing is completed. The harmony search analysis proved to be very efficient in choosing the best wavelet coefficient depending upon the structure of the signal. The extracted features are assigned towards corresponding classes and randomly divided as training and test data for the purpose of evaluation of the classifier. K-NN is used to classify the fault conditions of PV inverters into normal and faulty status. A five-fold cross validation is performed to measure the performance of the classifier with the input data. On validation, the developed approach depicted a training accuracy of 96.1 percent. Further, critically analysis is established form component-level information-based guidance for ranking failure mechanism.