A 3-Tier Fog-Cloud Orchestration Framework for Detecting Distributed Stealth Cyber-Attacks on Smart Meters Using Contrastive Learning and Public-Private Hybrid Blockchains

• Abiodun Okunola

  Ladoke Akintola University Technology

  Author

Smart meters within advanced metering infrastructure (AMI) remain vulnerable to stealthy false data injection attacks that evade conventional threshold-based detection due to their gradual, low-magnitude deviations from normal consumption patterns. Existing detection frameworks suffer from three critical limitations: reliance on labeled attack data that is scarce in operational settings, centralized processing architectures that introduce latency incompatible with real-time grid protection, and absence of tamper-evident audit trails for detected anomalies. This study designs and validates a three-tier fog-cloud orchestration framework that integrates contrastive learning for unsupervised stealth attack detection with a public-private hybrid blockchain layer for immutable anomaly logging. The framework employs federated self-supervised contrastive learning at fog nodes to learn representations of normal smart meter behavior without requiring labeled attack samples, while a permissioned blockchain consortium records detection events with cryptographic verification. Experimental validation on a smart meter electricity consumption dataset demonstrates that the contrastive learning detector achieves 89.4% accuracy, 91.2% precision, and 87.6% recall, outperforming autoencoder-based baselines by 12.3 percentage points while maintaining sub-100ms inference latency at fog nodes. The hybrid blockchain layer achieves 99.7% tamper-detection probability with 94.1% storage overhead reduction compared to full on-chain logging. The framework addresses the dual challenge of detecting distributed stealth attacks in unlabeled data streams while providing verifiable, privacy-preserving audit capabilities suitable for regulatory compliance.

• pdf

Copyright (c) 2026 Abiodun Okunola (Author)

This work is licensed under a Creative Commons Attribution 4.0 International License.