Enhanced Neural Network Method-Based Multiscale PCA for Fault Diagnosis: Application to Grid-Connected PV Systems

In this work, an effective Fault Detection and Diagnosis (FDD) strategy designed to increase the performance accuracy of fault diagnosis in grid-connected photovoltaic (GCPV) systems is developed. The evolved approach threefold: first, a pre-processing training dataset applied using multiscale scheme that decomposes data at multiple scales high-pass/low-pass filters separate noise from informative attributes prevent stochastic samples. Second, principal component analysis (PCA) technique newly obtained select, extract, preserve only more relevant, informative, uncorrelated attributes; finally, distinguish between diverse conditions, extracted are utilized train NNs classifiers. study, effort made take into consideration all potential frequent faults might occur PV systems. Thus, twenty-one faulty scenarios (line-to-line, line-to-ground, connectivity faults, can affect normal operation bay-pass diodes) have been introduced treated different levels locations; each scenario comprises various including occurrence simple PV1 array, PV2 PV1, PV2, mixed both arrays, order ensure complete global analysis, thereby reducing loss generated energy maintaining reliability efficiency such outcomes demonstrate proposed not achieves good accuracies but also reduces runtimes during process by avoiding noisy data, removing irrelevant correlated samples original dataset.