GridXAI-diffusion an adaptive graph intelligence framework for interpretable data integrity analysis in smart grid cybersecurity systems

Smart-grid cybersecurity is increasingly challenged by sophisticated cyberattacks, evolving operational behaviors, and complex cyber-physical dependencies. Existing machine learning, deep learning, and graph-based detection methods remain limited in jointly modeling dynamic attack behaviors, spatio-temporal dependencies, attack-data scarcity, and interpretable security decisions. This study proposes GridXAI-Diffusion, an adaptive and interpretable graph intelligence framework for smart-grid cyberattack detection and data-integrity analysis. Evaluation was conducted using the Kaggle Smart Grid Security Dataset containing 1,000 samples and 24 features representing operational signals, communication behaviors, extracted characteristics, threat indicators, and attack labels. The framework integrates diffusion-based cyberattack scenario generation, dynamic cyber-physical graph construction, spatio-temporal graph transformer learning, contrastive representation learning, and Graph Temporal Causal Attention Explainability (GTCAE). The principal novelty is the coordinated integration of these established techniques into a unified cybersecurity architecture, where each component addresses attack diversity, topology-aware representation, temporal evolution, robustness, or interpretability. Experimental results achieve 96.18% accuracy, 97.04% precision, 95.71% recall, and 96.82% F1-score, outperforming the evaluated baseline approaches. Ablation analysis further confirms complementary contributions from the integrated components. Overall, GridXAI-Diffusion provides an adaptive and interpretable framework for robust cyberattack detection and data-integrity assessment under evolving smart-grid conditions.