Search

Includes bibliographical references. Number of sources in the bibliography: 283. Thesis (Ph. D.) -- University of Cyprus, Faculty of Engineering, Department of Electrical and Computer Engineering, 2022. The University of Cyprus Library holds the printed form of the thesis. Κατά την τελευταία δεκαετία, η έρευνα στον τομέα των φωτοβολταϊκών (ΦΒ) συστημάτων έχει επικεντρωθεί σε μεθόδους που αποσκοπούν στην αύξηση της αξιοπιστίας και της απόδοσης των ΦΒ εγκαταστάσεων. Αρκετή έρευνα έχει διεξαχθεί τα τελευταία χρόνια στον τομέα αυτό με σκοπό την μείωση του σταθμισμένου κόστους ηλεκτρικής ενέργειας και την εγγυημένη απόδοση του συστήματος κατά τη διάρκεια της ζωής του, μέσω της συνεχούς και σταθερής παρακολούθησης των συστημάτων αυτών. Κύρια πρόκληση στην προσπάθεια διασφάλισης της βέλτιστης λειτουργίας των ΦΒ συστημάτων, είναι η διατήρηση της αξιοπιστίας μέσω της αυτόματης αναγνώρισης/διάγνωσης και ταξινόμησης σφαλμάτων σε διάφορους τύπους (π.χ. απώλειες απόδοσης, μηχανισμούς υποβάθμισης και αστοχίες). Η ικανότητα διάγνωσης σφαλμάτων σε πρώιμο στάδιο είναι επίσης ζωτικής σημασίας για τη βελτιστοποίηση της προληπτικής και διορθωτικής συντήρησης ΦΒ συστημάτων. Η παρούσα διδακτορική διατριβή προτείνει την ανάπτυξη αλγορίθμων και μεθοδολογιών για ένα σύστημα παρακολούθησης (το οποίο θα επεξεργάζεται και θα αναλύει δεδομένα σε πραγματικό χρόνο) που έχει ως στόχο την προληπτική και διορθωτική συντήρηση ΦΒ συστημάτων. Οι προτεινόμενοι αλγόριθμοι και οι καινοτόμες μεθοδολογίες (για αξιόπιστη διάγνωση σφαλμάτων και έγκαιρη επίλυση βλαβών λειτουργίας) μπορούν να ενσωματωθούν στο σύστημα παρακολούθησης για να διασφαλιστεί η βέλτιστη απόδοση και λειτουργία του ΦΒ συστήματος. Για τον εντοπισμό ΦΒ σφαλμάτων, οι καταγεγραμμένες μετρήσεις ελέγχονται αρχικά για ελλιπείς και λανθασμένες τιμές. Ακολούθως διάφοροι αλγόριθμοι (βασισμένοι σε τεχνικές στατιστικής ανάλυσης και μηχανικής μάθησης) αναπτύχθηκαν με σκοπό να: α) προβλέπουν την απόδοση του ΦΒ συστήματος, β) εντοπίζουν τις διάφορες αστοχίες, τους διάφορους μηχανισμούς υποβάθμισης και τις απώλειες απόδοσης, γ) ταξινομούν τα διάφορα σφάλματα σε κατηγορίες, δ) διακρίνουν τις βλάβες σε αναστρέψιμες και μη αναστρέψιμες και τέλος ε) παρέχουν εισηγήσεις/προτάσεις διορθωτικών μέτρων (λαμβάνοντας υπόψη οικονομικά κριτήρια) για την επίλυση των βλαβών/σφαλμάτων. Οι προτεινόμενες μεθοδολογίες εκτελούν συγκρίσεις μεταξύ των μετρούμενων και των προσομοιωμένων ηλεκτρικών παραμέτρων και χρησιμοποιούν αλγορίθμους μηχανικής μάθησης και στατιστικής για να εντοπίσουν και να κατηγοριοποιήσουν τα σφάλματα στο ΦΒ σύστημα. Η αποτελεσματικότητα του προτεινόμενου συστήματος για τη διάγνωση απωλειών απόδοσης και σφαλμάτων επαληθεύτηκε πειραματικά σε υφιστάμενα ΦΒ συστήματα (μικρής και μεγάλης κλίμακας) που είναι εγκατεστημένα στην Κύπρο, στην Ελλάδα και στις ΗΠΑ. Αυτές οι καινοτόμες μεθοδολογίες και οι αλγόριθμοι μπορούν να ενσωματωθούν σε υφιστάμενες ΦΒ εγκαταστάσεις για αυτόματη αναγνώριση σφαλμάτων, χωρίς να διαταράσσεται η κανονική λειτουργία του συστήματος. Over the last decade, research on photovoltaic (PV) systems has shifted from a race for the highest cell efficiency to the increase of the lifetime energy yield. Presently, a major part of the research concerns with the reliability of the installations and the guaranteed lifetime performance through constant, solid, and traceable PV plant monitoring of installed systems, hence directly impacting positively investment cost, levelized cost of electricity (LCOE), and in general PV competitiveness. In this sense, a main challenge in the quest for ensuring quality of operation for grid-connected PV systems, is to safeguard reliability and optimal performance by identifying and accurately quantifying the factors behind the various failures and performance loss mechanisms. The ability to diagnose (detect and classify) faults at an early stage is also crucial for optimising the maintenance strategies and reducing the operation and maintenance (O&M) costs. The present doctoral study proposes the development of algorithms for PV monitoring applications. In this context, innovative methodologies/routines and algorithms are developed to significantly improve the performance of PV systems through the early and reliable diagnosis of faults (i.e., failures and trend-based performance losses). To detect faults in PV systems, the recorded measurements are first thoroughly checked for consistency, missing, erroneous and outlying values. After checking and correcting the acquired measurements, different data-driven algorithms (based on statistical analysis and machine learning techniques) are developed to: a) predict the performance of the PV system under investigation, b) detect commonly occurred PV failures and performance losses, c) quantify the fault incidents and the root causes, d) distinguish reversible from irreversible fault mechanisms, and finally, e) provide recommendations of actionable O&M decisions – mitigation actions (e.g., suggestions of corrective actions) to resolve the detected incidents based on fault criticality evaluation and financial metrics. The proposed routines perform comparisons between the measured and simulated electrical parameters and utilise statistical (i.e., change-point) and machine learning algorithms to first identify the fault and then quantify the root cause. The effectiveness of the proposed algorithms and routines for diagnosing accurately fault occurrences was demonstrated experimentally on test PV systems (ranging from small- to large-scale) installed in Cyprus, Greece and US. Finally, the derived data-driven functionalities can then be embedded into existing PV monitoring systems for real-time performance monitoring and/or when mass analysing data from PV power plants. This enables remote and automated diagnosis of PV underperformance issues and the provision of recommendations for fault events resolution (by performing O&M activities), without affecting the overall operation of the PV system.

Dataset from FOSS power plant in Cyprus. Cell type: mono-C-Si.

The IEA PVPS Task 13 group, experts who focus on photovoltaic performance, operation, and reliability from several leading R&D centers, universities, and industrial companies, is developing a framework for the calculation of performance loss rates of a large number of commercial and research photovoltaic (PV) power plants and their related weather data coming across various climatic zones. The general steps to calculate the performance loss rate are (i) input data cleaning and grading; (ii) data filtering; (iii) performance metric selection, corrections, and aggregation; and finally, (iv) application of a statistical modeling method to determine the performance loss rate value. In this study, several high-quality power and irradiance datasets have been shared, and the participants of the study were asked to calculate the performance loss rate of each individual system using their preferred methodologies. The data are used for benchmarking activities and to define capabilities and uncertainties of all the various methods. The combination of data filtering, metrics (performance ratio or power based), and statistical modeling methods are benchmarked in terms of (i) their deviation from the average value and (ii) their uncertainty, standard error, and confidence intervals. It was observed that careful data filtering is an essential foundation for reliable performance loss rate calculations. Furthermore, the selection of the calculation steps filter/metric/statistical method is highly dependent on one another, and the steps should not be assessed individually.

Photovoltaic (PV) power prediction is important for monitoring the performance of PV plants. The scope of this work is to develop a methodology for deriving an optimized location and technology independent machine learning (ML) model for power prediction. The prediction accuracy results demonstrated that the performance of the ML model was primarily affected by the dataset split method. In particular, for a 70:30 % train and test set approach, the ML model achieved a normalized root mean square error (nRMSE) of 0.88 % when using randomly selected samples compared to 0.94 % when using continuous samples. The accuracy of the developed model was also affected by the duration of the train set. For a random 70:30 % train and test set approach, the constructed ML topology achieved a nRMSE of 0.88 %, while when the dataset was split into a 30:30 % portion, the nRMSE was 0.95 %. Moreover, when low irradiance conditions were filtered out and 70 % of the entire dataset was randomly chosen for model training, a nRMSE of 1.41 % was obtained demonstrating that the model's accuracy was not improved. Finally, for a random 10:30 % train and test set approach, the FNNN achieved the lowest nRMSE of 1.10 % when the model was trained using the prevailing irradiance classes. 1270 1275

In this work, the prediction performance of the mechanistic performance model (MPM) and a machine learning Feed Forward Neural Network (FFNN), was evaluated using yearly datasets, containing sixty-minute average and instantaneous field measurements obtained from the outdoor test sites in Nicosia, Cyprus and in Arizona, USA, respectively. The model exhibiting the lowest prediction error was derived based on different model training conditions. The obtained results demonstrated that both models provide good predictive quality using both instantaneous and average measurements. The performance of the models was strongly dependent on the duration of the train set, since for a random 70:30 % train and test set split using the yearly dataset from the GI OTF, a mean absolute percentage error (MAPE) of 1.95 % and 1.55 % was obtained for the MPM and FFNN, respectively. Alternatively, for a random 30:30 % train and test set data partition, the MPM and the FFNN achieved a MAPE of 2.03 % and 1.67 %, respectively. Moreover, by applying a medium irradiance condition filter (GI>0.3 kW/m 2) to the dataset during a random 70:30 % train and test set approach, the MPM and the FFNN achieved a MAPE of 1.77 % and 1.37 %, respectively, demonstrating that the predictive accuracy of the models was enhanced by data filtering. Additionally, the MPM and the FFNN achieved the lowest MAPE of 2.05 % and 2.11 % when the train set (10 % of the entire dataset) contained 80 % of clear and 20 % of variable and diffuse measurements (same amount of weather type measurements as the amount of the irradiance profile classes of the location). Finally, for accurate predictions a random train and test approach should be utilized, the training process should be performed using the greatest possible amount of train data samples and by simultaneously applying an irradiance condition filter.

Accurate and consistent performance assessment of photovoltaic (PV) systems with the use of advanced failure diagnostic tools is essential to safeguard high levels of production. The scope of this work is to propose a failure diagnostic approach capable of diagnosing short-and open-circuited PV modules in grid-connected PV systems. The developed failure diagnosis approach comprises of a failure detection and a classification stage and requires as input features the acquired in-plane global irradiance, ambient and module temperature and PV operational electrical measurements. In particular, the failure detection stage was based on a statistical comparative assessment between the measured and simulated dc power, while the failure classification procedure was based on the ratio between the measured and simulated dc voltage and current. The diagnostic approach was validated experimentally using field measurements of a PV system installed in Cyprus. The obtained results demonstrated that the developed methodology can detect and classify short-and open-circuit fault conditions in PV systems. More specifically, the failure diagnosis rates, obtained when emulating 100 failure signatures imputed to the yearly acquired performance time series data of the test system, were 98.6 % and 99.8 % for short-and open-circuited PV module failures, respectively. 739 744

35th European Photovoltaic Solar Energy Conference and Exhibition; 1548-1553, Real-time identification of failures in photovoltaic (PV) systems is crucial for achieving reactive maintenance schemes that, in turn, will increase the system reliability and guarantee the lifetime output. Following this line, failure detection routines (FDRs) that operate on acquired data-sets of grid-connected PV systems were developed to diagnose the occurrence of failures. The developed FDRs comprise of a failure detection and a classification stage. The detection stage was based on the comparison between the measured and predicted DC power production against set threshold levels (TL). The classification stage was based on data-driven algorithms, which were used to post-process the detected failure patterns through the application of the developed decision trees (DT), k-nearest neighbours (k-NN), support vector machine (SVM) and fuzzy inference systems (FIS). The experimental results showed that the FDRs were capable of detecting all the different types of failures (open- and short-circuited PV module, inverter shutdown, shorted bypass diode and partial shading) that were introduced to the test-bench PV system. Finally, amongst the investigated models, the k-NN model achieved the highest average true negative rate of 91%, when classifying each type of failure used for benchmarking.

Accurate identification of failures in photovoltaic (PV) systems, which can result in energy loss or even serious safety issues, is crucial for ensuring reliability of the installations and the guaranteed lifetime output. The scope of this paper is to present the development of failure detection routines (FDRs) that operate on acquired data-sets of grid-connected PV systems in order to diagnose the occurrence of failures. The developed FDRs comprise of a failure detection and classification stage. Specifically, the failure detection stage was based on a comparative statistical approach between the measured and simulated electrical measurements. In parallel, fuzzy logic inference was performed in order to analyse the failure pattern and classify accurately the occurred fault. The fuzzy rule based classification system (FRBCS) models were constructed for each failure through a supervisory learning process, trained with continuous samples split in a 70:30 % train and test set approach of acquired data-sets which included the feature patterns exhibited during normal and faulty operation. The results obtained by emulating three failure patterns (partial shading, inverter shutdown and bypass diode failure), showed that the developed FDRs were capable of detecting accurately the faults upon their occurrence by signifying detectable discrepancies through the daily statistical comparisons of the measured and simulated electrical parameters. Finally, the developed classification models showed high accuracy of classifying each failure occurrence within the test set used for benchmarking. Specifically, the success rate obtained with the FRBCS models was 100 %, 96.9 % and 96.53 %, when classifying the inverter shutdown failure, bypass diode fault and partial shading, respectively. 1 6

Accurate and consistent performance assessment of photovoltaic (PV) systems with the use of advanced failure diagnostic tools is essential to safeguard high levels of production. The scope of this work is to propose a failure diagnostic approach capable of diagnosing short-and open-circuited PV modules in grid-connected PV systems. The developed failure diagnosis approach comprises of a failure detection and a classification stage and requires as input features the acquired in-plane global irradiance, ambient and module temperature and PV operational electrical measurements. In particular, the failure detection stage was based on a statistical comparative assessment between the measured and simulated dc power, while the failure classification procedure was based on the ratio between the measured and simulated dc voltage and current. The diagnostic approach was validated experimentally using field measurements of a PV system installed in Cyprus. The obtained results demonstrated that the developed methodology can detect and classify short-and open-circuit fault conditions in PV systems. More specifically, the failure diagnosis rates, obtained when emulating 100 failure signatures imputed to the yearly acquired performance time series data of the test system, were 98.6 % and 99.8 % for short-and open-circuited PV module failures, respectively. 739 744

Potential Induced Degradation (PID) is expected to become more frequent with the increased system voltage of photovoltaic (PV) systems. Thus, a PID detection method is required to detect PID at an early stage in order to increase the reliability of PV systems and the lifetime of PV modules. In this paper, a relation between the shunt resistance and the power loss of a PV cell due to PID is derived for the development of a detection model from cell to module level. The results indicate a logarithmic relationship. In addition, a shunt resistance threshold, at which significant power loss starts to appear, is derived. The threshold is examined for various power loss levels and temperatures. The threshold was found to decrease with temperature with a near linear behaviour. 1950 1954

In this paper, the impact of missing data and the robustness of commonly used statistical techniques to calculate the annual degradation rate of crystalline-Silicon (c-Si) photovoltaic (PV) systems was analyzed. In addition, the performance of different imputation techniques was assessed in order to develop an optimal methodology for treating missing data for degradation rate estimation studies. The results obtained clearly demonstrate that the application of the different statistical methods to estimate the annual degradation rate is sensitive to the amount of missing data, since all the statistical methods underestimated the degradation rate consistently with the increasing level of missing data. In addition, the application of the Seasonal Decomposition (Seas) technique yielded robust annual degradation rate estimates since for a level of 40 % of missing data, the absolute percentage error (APE) of the annual degradation rate estimated with all statistical techniques, was less than 7.5 %. Finally, Classical Seasonal Decomposition (CSD) was shown to be the most robust technique for estimating the degradation rate when imputation was applied, while Autoregressive Integrated Moving Average (ARIMA) was the most successful technique in providing robust degradation rate estimates when not applying any imputation. 1954 1958