الفهرس 
INTRODUCTIONOnly 14 pages are availabe for public view
 |  | from | 151 |  |  |
| | | from | 151 | | |
Abstract
Photovoltaic (PV) systems represent one of the most interesting renewable energy technologies thanks to the flexibility of modular PV technology, which provides a wide variety of electrical needs. Despite its environmental benefits and other desirable features, the PV industry is still facing challenges such as capital expenditures, its colossal reliance on environmental conditions, and the risk of faults. Especially for PV arrays (Direct Current (DC)-Side of the PV system), due to the current-limiting nature and nonlinear output characteristics of PV arrays, faults in PV arrays may not be detected. Moreover, the conventional series-parallel PV configurations increase voltage and current ratings, leading to a higher risk of large fault currents or DC arcs. This dissertation presents the challenges and limitations of existing fault detection and protection solutions in PV arrays. It is shown that the PV array faults may not be detectable by conventional protection devices under certain conditions. Thus, the faults can remain hidden in the PV system, resulting in possible dangers including sub-optimal performance, DC arcing, fire hazards, etc. The success of artificial intelligence, especially Machine Learning Techniques (MLTs) in solving problems in diverse domains motivated the research community to extend the MLTs capability, to address the various limitations of existing fault detection and protection solutions, thanks to its capability to handle the nonlinear nature of the PV arrays. In this trend, this dissertation proposes a Fault Identification strategy based on a Supervised machine Learning approach “FISL”, for the early detection and diagnosis of faults in the DC-Side of PV systems. On the other hand, for alleviating the resource-intensive process and overcoming standard supervised machine learning algorithms limitations in terms of requiring vast computation time for the training process, a large amount of labeled data, and not having the self-updating capability. This dissertation also proposes an innovative fault identification strategy that combines the Ensemble Learning concept in conjunction with a Semi-supervised Self-Training method (ELSST), to identify faults in the DC-Side of PV systems. Due to the self-training “learning” ability of the proposed ELSST strategy, the labeled set can be updated over time even with weather changes or PV arrays degrade, increasing the PV’s safety and reliability. The effectiveness of the proposed strategies “FISL” and “ELSST” are verified by the simulation and experimental results. The findings showed that the proposed strategies are able to recognize fault states accurately under various solar irradiance levels and hybrid faults as well, which, in turn, enables PV operators to expedite the system restoration procedure, promising to elevate the performance and robustness of PV systems.