Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Scalable and Robust Physics-Informed Graph Neural Networks for Water Distribution Systems

Inaam Ashraf, André Artelt, Barbara Hammer

TL;DR

This work tackles hydraulic state estimation in water distribution systems under varying demands and pipe attributes by introducing SPI-GNN, a physics-informed graph neural network framework that couples a local GNN (f1) with a global physics-inspired solver (f2). A two-phase training scheme and physics-preserving normalization ensure stable training and adherence to hydraulic laws, even as networks scale and input features become out-of-distribution. The approach integrates complex components like pumps and PRVs, demonstrates robustness on larger WDSs, and achieves meaningful speed-ups over traditional hydraulic simulators and prior DL surrogates, addressing the simulation-to-real gap for planning and rehabilitation. The results indicate strong accuracy, generalization to unseen input distributions, and practical implications for rapid scenario analysis in water infrastructure management, with clear avenues for temporal modeling and broader tasks in future work.

Abstract

Water distribution systems (WDSs) are an important part of critical infrastructure becoming increasingly significant in the face of climate change and urban population growth. We propose a robust and scalable surrogate deep learning (DL) model to enable efficient planning, expansion, and rehabilitation of WDSs. Our approach incorporates an improved graph neural network architecture, an adapted physics-informed algorithm, an innovative training scheme, and a physics-preserving data normalization method. Evaluation results on a number of WDSs demonstrate that our model outperforms the current state-of-the-art DL model. Moreover, our method allows us to scale the model to bigger and more realistic WDSs. Furthermore, our approach makes the model more robust to out-of-distribution input features (demands, pipe diameters). Hence, our proposed method constitutes a significant step towards bridging the simulation-to-real gap in the use of artificial intelligence for WDSs.

Scalable and Robust Physics-Informed Graph Neural Networks for Water Distribution Systems

TL;DR

This work tackles hydraulic state estimation in water distribution systems under varying demands and pipe attributes by introducing SPI-GNN, a physics-informed graph neural network framework that couples a local GNN (f1) with a global physics-inspired solver (f2). A two-phase training scheme and physics-preserving normalization ensure stable training and adherence to hydraulic laws, even as networks scale and input features become out-of-distribution. The approach integrates complex components like pumps and PRVs, demonstrates robustness on larger WDSs, and achieves meaningful speed-ups over traditional hydraulic simulators and prior DL surrogates, addressing the simulation-to-real gap for planning and rehabilitation. The results indicate strong accuracy, generalization to unseen input distributions, and practical implications for rapid scenario analysis in water infrastructure management, with clear avenues for temporal modeling and broader tasks in future work.

Abstract

Water distribution systems (WDSs) are an important part of critical infrastructure becoming increasingly significant in the face of climate change and urban population growth. We propose a robust and scalable surrogate deep learning (DL) model to enable efficient planning, expansion, and rehabilitation of WDSs. Our approach incorporates an improved graph neural network architecture, an adapted physics-informed algorithm, an innovative training scheme, and a physics-preserving data normalization method. Evaluation results on a number of WDSs demonstrate that our model outperforms the current state-of-the-art DL model. Moreover, our method allows us to scale the model to bigger and more realistic WDSs. Furthermore, our approach makes the model more robust to out-of-distribution input features (demands, pipe diameters). Hence, our proposed method constitutes a significant step towards bridging the simulation-to-real gap in the use of artificial intelligence for WDSs.

Paper Structure

This paper contains 20 sections, 31 equations, 3 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Methodology
  4. Model Architecture
  5. Local Learnable GNN-Model
  6. Global Physics-Informed Algorithm
  7. Training Scheme
  8. Objective Function
  9. Data Normalization
  10. Experiments
  11. Datasets
  12. Eradicating Exploding Gradients
  13. Training Setup
  14. Results and Analysis
  15. Evaluation on Test Dataset
  16. ...and 5 more sections

Figures (3)

  • Figure 1: Architecture of the model incorporating a two-phase training scheme.
  • Figure 2: Robustness of SPI-GNN to change in demands.
  • Figure 3: Robustness of SPI-GNN to change in pipe diameters.