Integrated phaselet-neural network based approach to fault detection & localization in photovoltaic arrays

Dolati, Erfan; Ghaffarzadeh, Navid

doi:10.22034/a-see.2026.2090932.1026

Integrated phaselet-neural network based approach to fault detection & localization in photovoltaic arrays

Document Type : Original Article

Authors

Erfan Dolati ¹

navid ghaffarzadeh ²

¹ Imam Khomeini International University

² Imam Khomeini International university

10.22034/a-see.2026.2090932.1026

Abstract

To ensure reliable operation and prevent energy loss, equipment damage, or fire hazards in PV systems, timely and accurate fault detection and localization are essential. Conventional methods relying solely on total current or power often fail to distinguish faults that produce similar power drops — for instance, a line to ground fault on the first panel of different substrings may result in identical total power reduction. To overcome this limitation, this paper proposes a high-accuracy fault diagnosis method based on the Phaselet Transform of the instantaneous DC power signal and individual substring currents. A feedforward neural network is trained on several distinct scenarios (including healthy operation and various line to line and line to ground faults) and achieves an outstanding accuracy of 99.85% on the test dataset. For rare ambiguous cases where the network output does not uniquely identify a single fault location, a thermal imaging camera is suggested as a cost effective secondary verification tool.

Keywords

Fault detection

Fault Localization

photovoltaic systems

Phaselet

Neural network

Subjects