Finding Faults in PV Systems: Supervised and Unsupervised Dictionary Learning With SSTDR

Machine Learning-Based Condition Monitoring for PV Systems: State of the Art and Future Prospects

“…However, unlike the use of ensemble learning, the authors chose to trigger the learning process on multiple classifiers, namely, SVM, NB, and Logistic Regression (LR), and selected only the best after completion of the evaluation process. Edun et al [67] suggested a supervised and unsupervised dictionary learning approach to detect disconnect (open-circuit) faults in PV modules. Their main contribution involved the use of sparse coding algorithms, such as Singular Value Decomposition (SVD), to help locate and identify symptoms of faults in driven samples.…”

A SSTDR Methodology, Implementations, and Challenges

“…Reflectometry is a nondestructive technique to remotely detect, locate, and characterize useful information in physical systems. It has been used extensively in electrical systems to locate electrical “opens” or “shorts” [ 1 , 2 , 3 , 4 , 5 , 6 ], locate cable insulation break down [ 7 ], detect electrical system degradation [ 8 , 9 , 10 , 11 ], locate ground and arc faults [ 11 , 12 , 13 , 14 ], locate damaged components in photovoltaic (PV) systems [ 12 , 15 , 16 , 17 , 18 ], remotely measure complex impedance [ 17 , 19 , 20 , 21 , 22 ], locate intermittent faults that only occur during normal operation [ 3 , 4 , 8 , 23 , 24 , 25 , 26 ], and locate faults in three phase systems [ 27 ]. In material science, reflectometry has been used to characterize nanostructures, optical fiber networks, waveguides, and dielectrics [ 28 , 29 , 30 , 31 ].…”

“…It has been used in airplane cabling where it has been able to detect intermittent faults lasting a few milliseconds [ 47 ]. Application has also been seen in power systems [ 27 , 57 , 58 ], PV systems [ 11 , 12 , 13 , 15 , 16 , 18 , 59 , 60 ], undersea cabling [ 61 ], and railroad systems [ 49 , 62 , 63 ]. It has been proposed that SSTDR could be used in battery health monitoring [ 64 ] and cable degradation monitoring [ 65 ].…”

“…These shapes provide information that represents the complex impedance, such as capacitance and inductance [ 22 ]. Advanced algorithms such as machine learning and dictionary matching techniques have been used to identify the complex loads [ 18 , 19 , 110 ].…”

“…This dictionary is then compared to measured data to determine the health of the system and location(s) of faults. The dictionary matching methods has been used to measure the capacitance of an end-terminated load [ 19 ] and disconnection fault location in a PV system [ 18 ].…”