PhysTrust code in Smart Grids

This repository accompanies the research paper entitled “PhysTrust: Physics-Anchored Byzantine-Robust Federated Learning for Detecting Stealthy False Data Injection in Smart Grids.” It provides the computational framework for investigating physics-anchored Byzantine-robust federated learning for the detection of stealthy false data injection (FDI) attacks in smart grids. PhysTrust addresses two major challenges in data-driven smart-grid security: the inability of conventional residual-based bad-data detection (BDD) to detect certain AC-stealthy FDI attacks, and the vulnerability of federated learning (FL) to malicious model updates submitted by authenticated but compromised participants. The proposed approach uses the physical network model available to the coordinating operator to generate a synthetic, labelled reference dataset through AC power-flow simulations. This physics-derived reference guides the filtering, clipping, and robust aggregation of client updates without requiring the centralisation of private operational measurements. Main objectives The software supports the experimental evaluation of: Detection of stealthy and non-stealthy false data injection attacks. Federated intrusion detection under heterogeneous operating conditions. Robustness against malicious client updates and Byzantine model-poisoning attacks. Physics-based trust anchoring without relying on a private operational root dataset. Sensitivity to network-model mismatch and structured client heterogeneity. Computational overhead associated with robust aggregation and secure model-update authentication.