A Federated Physics-Informed Neural Network (f-PINN) and Zero-Knowledge Blockchain Architecture for Real-Time Stealthy FDIA Detection in Edge-Computing Smart Meters

• Adaan Ahsun

  Covenant University

  Author

False Data Injection Attacks (FDIAs) targeting edge-computing smart meters represent a critical and evolving threat to smart grid stability, with stealthy variants specifically engineered to evade conventional bad data detection mechanisms while maintaining physical consistency with power system state estimation. Existing detection approaches predominantly rely on centralized data aggregation, which introduces privacy vulnerabilities, communication bottlenecks, and single points of failure, while physics-agnostic machine learning models remain vulnerable to adversarially crafted stealthy injections that respect power flow constraints. This study proposes a federated physics-informed neural network (f-PINN) integrated with a zero-knowledge blockchain architecture for real-time, privacy-preserving detection of stealthy FDIAs at the edge. The methodology combines physics-informed neural network training constrained by power flow equations with federated learning to enable collaborative model development without raw data exchange, while zero-knowledge proofs on a permissioned blockchain ensure detection integrity and tamper-evident auditability. Experimental validation on IEEE 14-bus and 118-bus systems demonstrates that the f-PINN framework achieves 89.4% detection accuracy for stealthy FDIAs, outperforming conventional physics-agnostic baselines by 12.7 percentage points. The architecture achieves sub-second inference latency suitable for real-time SCADA deployment while maintaining data sovereignty across distributed grid operators. The framework provides a replicable blueprint for deploying privacy-preserving, physics-aware cybersecurity in critical energy infrastructure.

• pdf

Copyright (c) 2026 Adaan Ahsun (Author)

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