Preprint: Fuzzy-GNN and PINN-based Modeling of Water Infrastructure under SCADA Blackout Conditions
Jun 10, 2025··
1 min read
Tetiana-Starovoyt
Graph-based fuzzy PINN structure for adaptive modeling of blackout zones in water distributionAbstract
This working paper introduces an integrated hybrid modeling pipeline combining fuzzy graph neural networks (F-GNN), physics-informed neural networks (PINNs), and multi-objective optimization for adaptive water network control. The model addresses uncertainties due to partial SCADA blackout and proposes mechanisms for inferring missing hydraulic data. The framework was evaluated using a modified Walkerton EPANET model and synthetic fault scenarios. The results highlight the effectiveness of integrating PINNs with fuzzy logic to maintain system observability and resilience during data loss.
Type
This preprint builds upon the results of our previous published work on GIS-ANFIS-based accident prediction in water networks.
Due to the critical nature of infrastructure data and the wartime conditions in Ukraine, full access to source code and datasets is restricted. Only general architecture and anonymized examples are shared publicly.
The proposed model includes:
- Fuzzy membership layers on top of graph node attributes
- Integration of PINN for physical constraint modeling (e.g., Darcy–Weisbach)
- Compensation mechanisms for missing SCADA sensor inputs
- NSGA-II multi-objective optimization for valve control
This approach is applicable in utility AI systems that must remain resilient even during cyber-physical or war-induced infrastructure disruptions.