Leveraging dense layer hybrid graph neural networks for managing overvoltage in PV-dominated distribution systems
Asif Gulraiz, Syed Sajjad Haider Zaidi, Haseeb Gulraiz, Bilal Khan, Bilal Muhammad Khan, Majid Ali, Baseem Khan, Baseem Khan
Abstract
• Addresses overvoltage challenges in PV networks using innovative Graph Neural Network approach. • Introduces a novel Hybrid Graph Neural Network with Dense Layer (HGNN-DL) for advanced power system management. • Demonstrates superior predictive performance with lowest Mean Absolute Error and highest R² value. • Validated using both synthetic IEEE datasets and real-world field measurements. • Outperforms existing deep learning models including DST-GNN, PI-TGN, DGFN, and HE-IGAN in predictive accuracy. The rapid integration of photovoltaic (PV) systems into electrical distribution networks poses significant challenges, particularly regarding the management of overvoltage issues when generation exceeds demand. Conventional mitigation methods often struggle to address the complex, dynamic nature of modern power systems. This study introduces a novel Hybrid Graph Neural Network with Dense Layer (HGNN-DL) approach to tackle these challenges, leveraging advanced machine learning techniques for accurate voltage prediction and grid management. By employing advanced graph neural network techniques, the research addresses the complicated graph-structured characteristics of modern power distribution networks with high PV penetration. Comprehensive validation using both synthetic IEEE datasets and real-world field measurements demonstrates the superior performance of the proposed method. Comparative analysis across multiple deep learning models reveals the HGNN-DL method achieved remarkable predictive accuracy, with the lowest Mean Absolute Error (0.15000), Mean Squared Error (0.00250), and Root Mean Square Error (0.00550), coupled with an exceptional R² value of 0.97000. These results not only highlight the potential of advanced graph neural network architectures but also provide a promising framework for more effective overvoltage mitigation in photovoltaic (PV)- dominated electrical grids.