pith. sign in

arxiv: 2604.10166 · v1 · submitted 2026-04-11 · 💻 cs.LG · cs.AI· cs.SY· eess.SY

Virtual Smart Metering in District Heating Networks via Heterogeneous Spatial-Temporal Graph Neural Networks

Keivan Faghih Niresi , Christian M{\o}ller Jensen , Carsten Skovmose Kalles{\o}e , Rafael Wisniewski , Olga Fink This is my paper

Pith reviewed 2026-05-10 15:44 UTC · model grok-4.3

classification 💻 cs.LG cs.AIcs.SYeess.SY
keywords virtual sensingdistrict heatinggraph neural networksspatial-temporal modelingsmart meteringheterogeneous networksthermal energy systems
0
0 comments X

The pith

A heterogeneous spatial-temporal graph neural network reconstructs missing pressure, flow and temperature readings in district heating networks from sparse sensors.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper proposes HSTGNN to build virtual smart meters that estimate unmeasured states by capturing the coupled spatial and temporal dependencies among hydraulic and thermal variables. District heating networks are usually under-instrumented, so physical sensors cannot provide enough coverage for efficient control and fault detection. The model uses separate branches that learn graph structures and dynamics tailored to flow, temperature and pressure, allowing joint modeling of cross-variable correlations that simpler methods overlook. This approach is tested on a new laboratory dataset collected under controlled conditions that mimic real operation. The authors show the model outperforms standard baselines, offering a practical route to higher observability without adding hardware.

Core claim

The HSTGNN incorporates the functional relationships inherent in district heating networks and employs dedicated branches to learn graph structures and temporal dynamics for flow, temperature, and pressure measurements, thereby enabling the joint modeling of cross-variable and spatial correlations. On the released laboratory dataset of synchronized high-resolution measurements, the model significantly outperforms existing baselines for virtual sensing.

What carries the argument

Heterogeneous spatial-temporal graph neural network (HSTGNN) whose dedicated branches separately process graph topology and time series for each physical quantity while sharing information across variables.

If this is right

  • Virtual sensing accuracy improves without requiring dense synchronized sensor data across the entire network.
  • Joint modeling of pressure-flow-temperature dependencies becomes possible under nonlinear and heterogeneous conditions.
  • The released laboratory dataset provides a public benchmark that enables direct comparison of future virtual sensing methods.
  • Higher observability supports data-driven control, predictive optimization, and early fault detection in thermal networks.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • If the model generalizes to field conditions, operators could reduce physical sensor density and associated maintenance costs while maintaining monitoring quality.
  • The same architecture could be adapted to other spatially distributed fluid networks such as water distribution or gas pipelines.
  • Integration of the learned graph structures with existing hydraulic simulation tools might increase robustness when sensor faults occur.

Load-bearing premise

The laboratory dataset collected under controlled conditions is representative of the heterogeneous topologies and operating regimes found in real district heating networks.

What would settle it

When the same HSTGNN is applied to measurements from an operating district heating network, its prediction errors for pressure, flow or temperature exceed those of simpler baselines by a large margin.

Figures

Figures reproduced from arXiv: 2604.10166 by Carsten Skovmose Kalles{\o}e, Christian M{\o}ller Jensen, Keivan Faghih Niresi, Olga Fink, Rafael Wisniewski.

Figure 1
Figure 1. Figure 1: Schematic layout of the district heating network under study. The tree-structured graph illustrates the symmetrical [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Schematic block diagram of the experimental topology. The system is connected in a tree structure, linking the heat [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Hardware implementation of the experimental setup at Aalborg SWIL. [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 5
Figure 5. Figure 5: This multi-branch design ensures that the heterogeneous nature of the sensors is explicitly respected [PITH_FULL_IMAGE:figures/full_fig_p010_5.png] view at source ↗
Figure 5
Figure 5. Figure 5: The Proposed HSTGNN Architecture for Virtual Sensing [PITH_FULL_IMAGE:figures/full_fig_p011_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Outlet temperature prediction for SM 1, Dataset 1. HSTGNN closely follows the true signal; GRU-GCN is second [PITH_FULL_IMAGE:figures/full_fig_p017_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Inlet temperature for SM 1, Dataset 2. HSTGNN is most accurate; temporal-only and graph-only models deviate [PITH_FULL_IMAGE:figures/full_fig_p017_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Inlet temperature for SM 2, Dataset 3. HSTGNN tracks the measured signal well; GRU-GCN captures general [PITH_FULL_IMAGE:figures/full_fig_p018_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Flow rate for SM 1, Dataset 4. HSTGNN and GRU-GCN perform best; graph-only models outperform temporal-only [PITH_FULL_IMAGE:figures/full_fig_p018_9.png] view at source ↗
read the original abstract

Intelligent operation of thermal energy networks aims to improve energy efficiency, reliability, and operational flexibility through data-driven control, predictive optimization, and early fault detection. Achieving these goals relies on sufficient observability, requiring continuous and well-distributed monitoring of thermal and hydraulic states. However, district heating systems are typically sparsely instrumented and frequently affected by sensor faults, limiting monitoring. Virtual sensing offers a cost-effective means to enhance observability, yet its development and validation remain limited in practice. Existing data-driven methods generally assume dense synchronized data, while analytical models rely on simplified hydraulic and thermal assumptions that may not adequately capture the behavior of heterogeneous network topologies. Consequently, modeling the coupled nonlinear dependencies between pressure, flow, and temperature under realistic operating conditions remains challenging. In addition, the lack of publicly available benchmark datasets hinders systematic comparison of virtual sensing approaches. To address these challenges, we propose a heterogeneous spatial-temporal graph neural network (HSTGNN) for constructing virtual smart heat meters. The model incorporates the functional relationships inherent in district heating networks and employs dedicated branches to learn graph structures and temporal dynamics for flow, temperature, and pressure measurements, thereby enabling the joint modeling of cross-variable and spatial correlations. To support further research, we introduce a controlled laboratory dataset collected at the Aalborg Smart Water Infrastructure Laboratory, providing synchronized high-resolution measurements representative of real operating conditions. Extensive experiments demonstrate that the proposed approach significantly outperforms existing baselines.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The manuscript proposes a heterogeneous spatial-temporal graph neural network (HSTGNN) for virtual smart metering in district heating networks. The model uses dedicated branches to learn graph structures and temporal dynamics separately for flow, temperature, and pressure, enabling joint modeling of cross-variable and spatial correlations. It introduces a new controlled laboratory dataset collected at the Aalborg Smart Water Infrastructure Laboratory with synchronized high-resolution measurements and claims that extensive experiments show HSTGNN significantly outperforms existing baselines for virtual sensing.

Significance. If the empirical results hold and generalize, the work could advance data-driven virtual sensing for sparsely instrumented thermal networks, supporting improved efficiency, reliability, and fault detection. The public release of the lab dataset is a clear strength that enables future benchmarking. However, the significance is limited by the exclusive reliance on a single controlled lab setting without demonstrated robustness to real-world heterogeneity.

major comments (2)
  1. [Experiments and Dataset] The central performance claim (HSTGNN significantly outperforms baselines) is demonstrated exclusively on the Aalborg laboratory dataset. Without explicit validation against field data, ablations on perturbed topologies, or tests under unmodeled disturbances and sensor faults, the reported margin may not support the broader claim of utility in heterogeneous district heating networks (see Experiments section and dataset description).
  2. [Abstract and §3 (Dataset)] The abstract asserts that the lab dataset provides measurements 'representative of real operating conditions,' but this is a load-bearing assumption for the outperformance claim. The manuscript should quantify how the controlled lab topology (pipe lengths, consumer dynamics, synchronization) maps to real networks or include sensitivity analyses to rule out regime-specific artifacts.
minor comments (2)
  1. [Abstract] The abstract states outperformance but provides no quantitative metrics, error bars, baseline details, or ablation results; the Experiments section should include these for reproducibility and to allow assessment of effect sizes.
  2. [Method] Clarify the exact functional relationships from district heating physics that are hard-coded versus learned in the heterogeneous graph construction.

Simulated Author's Rebuttal

2 responses · 1 unresolved

We thank the referee for the constructive comments on our manuscript. We have carefully considered the concerns regarding generalization from the laboratory dataset and have revised the paper to include additional experiments and clarifications. Below we respond point-by-point to the major comments.

read point-by-point responses

  1. Referee: [Experiments and Dataset] The central performance claim (HSTGNN significantly outperforms baselines) is demonstrated exclusively on the Aalborg laboratory dataset. Without explicit validation against field data, ablations on perturbed topologies, or tests under unmodeled disturbances and sensor faults, the reported margin may not support the broader claim of utility in heterogeneous district heating networks (see Experiments section and dataset description).

    Authors: We agree that exclusive reliance on a single laboratory dataset limits the strength of broader claims about utility in heterogeneous real-world networks. The controlled lab environment was chosen to enable precise, synchronized ground-truth measurements that are rarely available in operational systems. In the revised manuscript we have added: (i) ablations on perturbed topologies by systematically varying pipe lengths, diameters, and consumer load profiles within the lab setup; (ii) tests under simulated sensor faults (random dropouts and bias) and unmodeled disturbances (sudden valve changes). These new results show that the performance margin over baselines remains consistent. We have also expanded the discussion section to explicitly state that field-data validation is an important next step and that the public dataset release is intended to facilitate exactly such studies by the community. revision: partial

  2. Referee: [Abstract and §3 (Dataset)] The abstract asserts that the lab dataset provides measurements 'representative of real operating conditions,' but this is a load-bearing assumption for the outperformance claim. The manuscript should quantify how the controlled lab topology (pipe lengths, consumer dynamics, synchronization) maps to real networks or include sensitivity analyses to rule out regime-specific artifacts.

    Authors: We accept that the original phrasing overstated the representativeness. In the revised version we have: (i) softened the abstract claim to 'capturing key hydraulic and thermal dynamics observed in real district heating networks'; (ii) added a new quantitative mapping subsection in §3 that compares lab pipe lengths and diameters to typical values from Danish and Swedish district heating literature, consumer dynamics to standard residential/commercial load models, and synchronization accuracy to field sensor specifications; (iii) included sensitivity analyses varying flow rates, temperature differentials, and additive noise levels representative of field conditions. These analyses confirm that relative performance gains are stable across the tested regimes, reducing the risk of regime-specific artifacts. revision: yes

standing simulated objections not resolved
  • Direct validation against synchronized high-resolution field data from operational heterogeneous district heating networks, as no such public dataset was available to the authors.

Circularity Check

0 steps flagged

No significant circularity in derivation or claims

full rationale

The paper proposes HSTGNN as a modeling architecture for virtual sensing and evaluates it empirically on a newly collected laboratory dataset against external baselines. No mathematical derivation, prediction, or first-principles result reduces to its own inputs by construction. The architecture choices, dataset introduction, and performance comparisons are presented as design and experimental decisions rather than tautological redefinitions or self-referential fits. The central claims rest on held-out experimental results, making the work self-contained against the provided benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The central claim depends on the assumption that a graph representation plus variable-specific temporal branches can capture the coupled nonlinear physics without explicit hydraulic or thermal equations, and that the lab data distribution matches deployment conditions.

axioms (2)
  • domain assumption The functional relationships in district heating networks can be adequately learned from data without incorporating explicit physical constraints or simplified hydraulic models.
    Invoked when the authors state that analytical models rely on assumptions that may not capture heterogeneous topologies, and position the data-driven HSTGNN as the solution.
  • domain assumption Synchronized high-resolution laboratory measurements are sufficient to train and validate a model that generalizes to real district heating networks.
    Central to the validation strategy described in the abstract.

pith-pipeline@v0.9.0 · 5589 in / 1407 out tokens · 69403 ms · 2026-05-10T15:44:52.584077+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

35 extracted references · 35 canonical work pages · 2 internal anchors

  1. [1]

    Connolly, H

    D. Connolly, H. Lund, B. V. Mathiesen, S. Werner, B. Möller, U. Persson, T. Boermans, D. Trier, P. A. Østergaard, S. Nielsen, Heat roadmap europe: Combining district heating with heat savings to decarbonise the eu energy system, Energy policy 65 (2014) 475–489

    work page 2014

  2. [2]

    H. Lund, S. Werner, R. Wiltshire, S. Svendsen, J. E. Thorsen, F. Hvelplund, B. V. Mathiesen, 4th gen- eration district heating (4gdh): Integrating smart thermal grids into future sustainable energy systems, energy 68 (2014) 1–11

    work page 2014

  3. [3]

    Werner, International review of district heating and cooling, Energy 137 (2017) 617–631

    S. Werner, International review of district heating and cooling, Energy 137 (2017) 617–631. 20

    work page 2017

  4. [4]

    A. Hast, S. Syri, V. Lekavičius, A. Galinis, District heating in cities as a part of low-carbon energy system, Energy 152 (2018) 627–639

    work page 2018

  5. [5]

    Kadlec, B

    P. Kadlec, B. Gabrys, S. Strandt, Data-driven soft sensors in the process industry, Computers & chemical engineering 33 (4) (2009) 795–814

    work page 2009

  6. [6]

    Saloux, K

    E. Saloux, K. Zhang, Towards integration of virtual meters into building energy management systems: Development and assessment of thermal meters for cooling, Journal of Building Engineering 65 (2023) 105785

    work page 2023

  7. [7]

    S. Yoon, Y. Choi, J. Koo, Y. Hong, R. Kim, J. Kim, Virtual sensors for estimating district heating energy consumption under sensor absences in a residential building, Energies 13 (22) (2020) 6013

    work page 2020

  8. [8]

    Q. Sun, Z. Ge, A survey on deep learning for data-driven soft sensors, IEEE Transactions on Industrial Informatics 17 (9) (2021) 5853–5866

    work page 2021

  9. [9]

    O. Fink, I. Nejjar, V. Sharma, K. F. Niresi, H. Sun, H. Dong, C. Xu, A. Wei, A. Bizzi, R. Theiler, et al., From physics to machine learning and back: Part ii-learning and observational bias in phm, arXiv preprint arXiv:2509.21207 (2025)

    work page arXiv 2025

  10. [10]

    C. D. Petersen, R. Fraanje, B. S. Cazzolato, A. C. Zander, C. H. Hansen, A kalman filter approach to virtual sensing for active noise control, Mechanical Systems and Signal Processing 22 (2) (2008) 490–508

    work page 2008

  11. [11]

    K. F. Niresi, L. Kuhn, G. Frusque, O. Fink, Informed graph learning by domain knowledge injection and smooth graph signal representation, in: 2024 32nd European Signal Processing Conference (EUSIPCO), IEEE, 2024, pp. 2467–2471

    work page 2024

  12. [12]

    K. F. Niresi, H. Bissig, H. Baumann, O. Fink, Physics-enhanced graph neural networks for soft sensing in industrial internet of things, IEEE Internet of Things Journal (2024)

    work page 2024

  13. [13]

    M. Jin, H. Y. Koh, Q. Wen, D. Zambon, C. Alippi, G. I. Webb, I. King, S. Pan, A survey on graph neural networks for time series: Forecasting, classification, imputation, and anomaly detection, IEEE Transactions on Pattern Analysis and Machine Intelligence (2024)

    work page 2024

  14. [14]

    Liang, G

    Y. Liang, G. Huang, Z. Zhao, Joint demand prediction for multimodal systems: A multi-task multi- relational spatiotemporal graph neural network approach, Transportation research part C: emerging technologies 140 (2022) 103731

    work page 2022

  15. [15]

    Z. Li, L. Ye, X. Song, Y. Luo, M. Pei, K. Wang, Y. Yu, Y. Tang, Heterogeneous spatiotemporal graph convolution network for multi-modal wind-pv power collaborative prediction, IEEE Transactions on Power Systems 39 (4) (2023) 5591–5608

    work page 2023

  16. [16]

    Theiler, O

    R. Theiler, O. Fink, Heterogeneous graph neural networks for short-term state forecasting in power systems across domains and time scales: A hydroelectric power plant case study, arXiv preprint arXiv:2507.06694 (2025)

    work page internal anchor Pith review arXiv 2025

  17. [17]

    F.Terroso-Saenz, J.Morales-García, A.Munoz, Nationwideairpollutionforecastingwithheterogeneous graph neural networks, ACM Transactions on Intelligent Systems and Technology 15 (1) (2024) 1–19

    work page 2024

  18. [18]

    K.F.Niresi, I.Nejjar, O.Fink, Efficientunsuperviseddomainadaptationregressionforspatial-temporal sensor fusion, IEEE Internet of Things Journal (2025)

    work page 2025

  19. [19]

    M. Zhao, C. Taal, S. Baggerohr, O. Fink, Graph neural networks for virtual sensing in complex systems: Addressing heterogeneous temporal dynamics, Mechanical Systems and Signal Processing 230 (2025) 112544. 21

    work page 2025

  20. [20]

    K. F. Niresi, J. Qing, M. Zhao, O. Fink, Time-vertex machine learning for optimal sensor placement in temporal graph signals: Applications in structural health monitoring, Reliability Engineering & System Safety (2025) 112153

    work page 2025

  21. [21]

    Rodrigue, R

    D. Rodrigue, R. Boghetti, J. H. Kämpf, B. Pasdeloup, M. T. Mabrouk, P. Meyer, B. Lacarrière, Validation of two distinct simulation models of district heating networks: application to efficient looping analysis, in: Journal of Physics: Conference Series, Vol. 3140, IOP Publishing, 2025, p. 042021

    work page 2025

  22. [22]

    Boghetti, J

    R. Boghetti, J. H. Kämpf, A benchmark for the simulation of meshed district heating networks based on anonymised monitoring data, in: Journal of Physics: Conference Series, Vol. 2600, IOP Publishing, 2023, p. 022008

    work page 2023

  23. [23]

    T. N. Kipf, M. Welling, Semi-supervised classification with graph convolutional networks, in: Interna- tional Conference on Learning Representations, 2017

    work page 2017

  24. [24]

    Veličković, G

    P. Veličković, G. Cucurull, A. Casanova, A. Romero, P. Liò, Y. Bengio, Graph attention networks, in: International Conference on Learning Representations, 2018

    work page 2018

  25. [25]

    W. L. Hamilton, Graph representation learning, Morgan & Claypool Publishers, 2020

    work page 2020

  26. [26]

    J. Gao, B. Ribeiro, On the equivalence between temporal and static equivariant graph representations, in: International Conference on Machine Learning, PMLR, 2022, pp. 7052–7076

    work page 2022

  27. [27]

    Z. Wu, S. Pan, F. Chen, G. Long, C. Zhang, P. S. Yu, A comprehensive survey on graph neural networks, IEEE transactions on neural networks and learning systems 32 (1) (2020) 4–24

    work page 2020

  28. [28]

    Val Ledesma, R

    J. Val Ledesma, R. Wisniewski, C. S. Kallesøe, Smart water infrastructures laboratory: Reconfigurable test-beds for research in water infrastructures management, Water 13 (13) (2021) 1875

    work page 2021

  29. [29]

    A. Cini, D. Zambon, C. Alippi, Sparse graph learning from spatiotemporal time series, Journal of Machine Learning Research 24 (242) (2023) 1–36

    work page 2023

  30. [30]

    A. Cini, A. Jenkins, D. Mandic, C. Alippi, F. M. Bianchi, Relational conformal prediction for correlated time series, arXiv preprint arXiv:2502.09443 (2025)

    work page arXiv 2025

  31. [31]

    W. Kool, H. Van Hoof, M. Welling, Ancestral gumbel-top-k sampling for sampling without replacement, Journal of Machine Learning Research 21 (47) (2020) 1–36

    work page 2020

  32. [32]

    Estimating or Propagating Gradients Through Stochastic Neurons for Conditional Computation

    Y. Bengio, N. Léonard, A. Courville, Estimating or propagating gradients through stochastic neurons for conditional computation, arXiv preprint arXiv:1308.3432 (2013)

    work page internal anchor Pith review arXiv 2013

  33. [33]

    Atwood, D

    J. Atwood, D. Towsley, Diffusion-convolutional neural networks, Advances in neural information pro- cessing systems 29 (2016)

    work page 2016

  34. [34]

    Y. Li, R. Yu, C. Shahabi, Y. Liu, Diffusion convolutional recurrent neural network: Data-driven traffic forecasting, in: International Conference on Learning Representations, 2018

    work page 2018

  35. [35]

    Vaswani, N

    A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, Ł. Kaiser, I. Polosukhin, Attention is all you need, Advances in neural information processing systems 30 (2017). 22

    work page 2017