Project Abstract

The equivalent circuit models of the photovoltaic (PV) modules provide parametric information which grants researchers direct insights into the electrical properties and physical phenomena that transpire within the PV cells.

However, these models are rarely used for the online fault diagnosis of PV arrays due to their inherent complexities, i.e., the presence of multiple unknown parameters and exponential terms, and the transcendental nature of the model equations. Solving for the model entails a large amount of operating data point information.

The limited amount of information provided by on-site sensors meant that the existing PV fault diagnosis methods: (1) lacks the ability to distinguish concurrent faults; and (2) are unable to quantify the extent of detected faults. Both aspects are critical for accurate yield prediction and timely deployment of maintenance efforts. In recent years, the advent of smart inverters with I-V curve tracing capabilities has opened the opportunity to approach the analysis and identification of PV array faults from a new perspective.

This work proposes to exploit the wealth of information provided by the real-time PV array output characteristics, their corresponding model parameters, and the onsite sensor measurements to mitigate the abovementioned shortcomings. Through extensive investigation of the PV array faults and their effects on these key aspects, it is envisaged that a computationally efficient script-based PV model can be developed for the conception of a powerful semi-supervised learning based fault detection and diagnosis scheme.

The solution is expected to furnish both new and existing installations with the ability to reliably identify and quantify the severity of concurrent PV array faults.