Towards the Development of a Photovoltaic Array Fault Detection and Diagnosis (PVAFDD) System

The perpetual increment in energy demand continues to put pressure on the South African Economy. Independent Power Producers (IPPs) have been contracted to relieve the strain by supplementing energy production using solar photovoltaic (PV) technologies. These IPPs are paid per megawatt they produce and face stiff penalties should they fail to deliver on contractual obligations. Naturally solar PV plants are susceptible to numerous PV faults that could lead to a negative return on investment. It therefore makes economic sense to adopt mechanisms that can be able to detect, localize and diagnose PV faults when they occur within a solar PV system. There exists an extensive literature on how to detect and diagnose PV faults, however, localizing PV faults is still in its infancy. This paper proposes to cater to the needs of the IPPs by developing an intelligent PV Array Fault Detection and Diagnostics (PVAFDD) system capable of localizing PV faults which can be embedded into the Supervisory Control and Data Acquisition (SCADA) system used to manage and control such PV systems. The PVAFDD system is based on a machine learning (ML) model implemented using logistic regression algorithm. The ML model is trained using meteorological data ranging over a period of eight years in Alice, Eastern Cape. Using simulations driven by real-life data scenarios, we have been able to train, validate and test the PVAFDD system. When the PVAFDD system detects a fault, a cascade of real-time PVA tests is undertaken to localize the PV fault. The system then carries out PVA fault diagnostics and gives recommendations on the PV fault classification. Corrective measures can therefore be implemented on the affected PVA swiftly reducing the downtime of the PV plant, ergo proving to be a cost-effective measure that offers a competitive edge to IPPs using the PVAFDD system.