Deep Learning Based Faults Diagnosis in Grid-Connected Photovoltaic Systems

EasyChair Preprint 10410

Abstract

As a renewable energy source, the establishment of photovoltaic (PV) system has essentially expanded. In any case, due to the maturing impacts and external conditions, during operation, PV systems can incur failures. These failures may affect many system components, including converters, PV modules, and connecting lines, which could decrease system effectiveness and performance or even cause the system breakdown.  Thus, the fault detection and diagnosis (FDD) is an important aspect in high-efficiency grid-connected PV systems. Deep learning (DL) is used in the most well-known data-driven methodologies. The main benefit of DL algorithms for diagnosis is that they create a high-order, non-linear, and adaptive effort to memorize high-level highlights from PV data, the fault is then classified.  Therefore, a comparison of FDD-based DL approaches is presented in this article.  These methods include Long-Short Term Memory (LTM), Convolutional Neural Networks (CNN), and Neural Networks (NN). The implementation of the DL techniques-based fault diagnosis is done using an emulated Grid-Connected PV (GCPV) system.

To evaluate the effectiveness of the proposed approaches, we utilize data obtained from a healthy case, which are then injected with several fault scenarios in the DC side and AC side:  one fault in the PV sensor, two faults in the PV array level, this is about the DC side and in the other side there are the three-phase inverter fault and the grid external connection fault.

The proposed techniques achieved accuracy from 61.24% to 95.12%, and the models' performance is evaluated.

Keyphrases: Deep Learning (DL)., Fault Detection and Diagnosis (FDD), Grid Connected Photovoltaic System (GCPV)

@booklet{EasyChair:10410,
  author    = {Mansour Hajji},
  title     = {Deep Learning Based Faults Diagnosis in Grid-Connected Photovoltaic Systems},
  howpublished = {EasyChair Preprint 10410},
  year      = {EasyChair, 2023}}