Neural Network-Based Approach for Islanding Detection in a PV Grid-Connected System

Modern power networks are more dependent on smaller dispersed generating units as a result of the increased focus on renewable energy. The identification of islanding occurrences in these grids is more difficult than in conventional power grids, which rely on bigger centralised units. In grid-tied, PV systems, the difficulty in grid connection caused by unintentional islanding still presents a barrier. However, keeping the system connected to the power grid is crucial to ensuring high system stability and when an islanding condition occurs, the grid is de-energised to safeguard any personnel who may be on the grid trying to determine what caused the islanding. The three types of methods for islanding detection are communication-based, active, and passive. Active approaches use noise to estimate the size of the grid and measure the impact of that noise to identify islanding, which lowers the grid's quality of electricity. The implementation of communication-based techniques is constrained by their high cost and dependency on the communications infrastructure. Unlike existing techniques, the suggested method of Artificial Neural Network (ANN) depends on the modified negative sequence impedance values and phase angle jump values to properly identify every islanding situation with a 98.9% accuracy. Without using any feature extraction, these data are provided directly into the ANN, enabling quicker and more accurate decision-making. In order to accomplish faster and more accurate detection, the suggested technique is anticipated to combine the high accuracy of ANNs with the rapid change in instantaneous values. A 100 kW of PV integrated grid system is used for the study using MATLAB/Simulink (version 2018a).