[Paper Review] Optimal Sampling of Water Distribution Network Dynamics using Graph Fourier Transform

Zhuangkun Wei, Alessio Pagani+5 more|arXiv (Cornell University)|Apr 6, 2019

Water Systems and Optimization30 references4 citations

TL;DR

This paper proposes a novel Graph Fourier Transform (GFT)-based method for optimal sampling of water distribution network (WDN) dynamics by exploiting the low-rank structure of pollution spread data. By identifying a minimal set of critical junctions—as few as 30–40% of total junctions—using a data-driven GFT operator, the framework enables full recovery of network-wide dynamics, outperforming numerical optimization, compressed sensing, and graph-theoretic approaches in sample efficiency and theoretical guarantees.

ABSTRACT

Water Distribution Networks (WDNs) are critical infrastructures that ensure safe drinking water. One of the major threats is the accidental or intentional injection of pollutants. Data collection remains challenging in underground WDNs and in order to quantify its threat to end users, modeling pollutant spread with minimal sensor data is can important open challenge. Existing approaches using numerical optimisation suffer from scalability issues and lack detailed insight and performance guarantees. Applying general data-driven approaches such as compressed sensing (CS) offer limited improvements in sample node reduction. Graph theoretic approaches link topology (e.g. Laplacian spectra) to optimal sensing locations, it neglects the complex dynamics. In this work, we introduce a novel Graph Fourier Transform (GFT) that exploits the low-rank property to optimally sample junction nodes in WDNs. The proposed GFT allows us to fully recover the full network dynamics using a subset of data sampled at the identified nodes. The proposed GFT technique offers attractive improvements over existing numerical optimisation, compressed sensing, and graph theoretic approaches. Our results show that, on average, with nearly 30-40\% of the junctions monitored, we are able to fully recover the dynamics of the whole network. The framework is useful beyond the application of WDNs and can be applied to a variety of infrastructure sensing for digital twin modeling.

Motivation & Objective

• Address the challenge of minimizing sensor count in water distribution networks (WDNs) while ensuring full recovery of dynamic pollution spread.
• Overcome scalability and lack of theoretical guarantees in existing numerical optimization and compressed sensing approaches.
• Integrate network topology and fluid dynamics by developing a data-adaptive GFT that captures low-rank dynamics.
• Provide a theoretically grounded, scalable solution for sensor placement that ensures full network reconstruction from minimal sampling.
• Enable digital twin modeling for critical infrastructure by offering a generalizable framework beyond WDNs.

Proposed method

• Propose a data-driven Graph Fourier Transform (GFT) operator tailored to WDN dynamics, derived from the singular value decomposition (SVD) of the dynamic state matrix X.
• Formulate the GFT such that the signal X becomes rank-r and bandlimited in the frequency domain, with non-zero energy concentrated in the first r frequencies.
• Identify optimal sampling nodes as the first r rows of the GFT matrix, ensuring that sampling at these nodes allows full reconstruction via inverse GFT.
• Theoretical recovery is guaranteed when the number of sampled nodes |S| ≥ rank(X), leveraging the low-rank property of pollution dynamics.
• Use a sampling scheme based on the GFT frequency response, where the most informative nodes correspond to the dominant singular vectors of the data matrix.
• Validate the method using synthetic pollution dynamics on real WDN topologies, comparing against Laplacian-based graph methods and compressed sensing.

Experimental results

Research questions

• RQ1 Can a data-driven GFT be designed to optimally sample WDN junctions while ensuring full dynamic recovery with minimal sensors?
• RQ2 How does the proposed GFT-based sampling compare in performance and sample efficiency to numerical optimization, compressed sensing, and graph-theoretic methods?
• RQ3 To what extent does the low-rank structure of WDN dynamics enable reduced sampling without loss of reconstruction fidelity?
• RQ4 What is the theoretical minimum number of sampling nodes required for full recovery, and how does it relate to the rank of the dynamic state matrix?
• RQ5 Can the proposed framework be generalized beyond WDNs for digital twin modeling in other infrastructure systems?

Key findings

• The proposed method achieves full recovery of WDN dynamics using only 30–40% of junctions on average, significantly reducing sensor count compared to full monitoring.
• The minimum number of sampling nodes |S|min equals the rank of the dynamic state matrix X, achieving theoretical optimality.
• The method consistently achieves RMSE < 10⁻⁸ across 100 test cases, demonstrating robustness across diverse pollution dynamics.
• The required sampling bandwidth R_min is exactly equal to rank(X), outperforming both Laplacian-based and compressed sensing schemes.
• The data-driven GFT operator enables full reconstruction with fewer samples than compressed sensing and graph-theoretic methods, which rely on topology alone.
• The framework is generalizable and applicable to digital twin modeling in other infrastructure systems beyond WDNs.