XAI Enhanced GCNN-HSA Framework for Anomaly Detection in Smart Grids

Abstract

The integration of digital technologies enhances smart grid connectivity, dependability, and sustainability, but their growing complexity necessitates advanced, intelligent anomaly detection for secure operation. This study proposes a novel hybrid framework combining Graph Convolutional Neural Networks (GCNN) with the Harmony Search Algorithm (HSA) for robust anomaly detection in smart grids. HSA optimizes GCNN hyper parameters, significantly boosting detection accuracy and responsiveness. A key innovation is the integration of Explainable Artificial Intelligence (XAI) techniques, specifically SHAP and Grad-CAM, to render the model's decision-making transparent and interpretable. This allows stakeholders, including operators and analysts, to better understand, validate, and trust the model's predictions. Experimental evaluations on the IEC 60870-5-104 and public cyberattack datasets confirm the proposed GCNN-HSA framework's superior performance in accuracy, precision, recall, F1-score, and AUROC compared to conventional methods. The XAI components further enhance system usability and accountability. This research contributes a novel, high-performance, and inherently explainable anomaly detection framework, addressing both technical efficacy and operational transparency to foster more secure, reliable, and interpretable smart grid infrastructures