Optimal dynamic thermal plant control: A study and benchmark

Gundolf Haase; Stefano Coss; Thomas Grandits

REVIEW 4 major objections 5 minor 25 references

Optimal dynamic thermal plant control: A study and benchmark

T0 review · 4 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read This paper argues that district heating plant schedules can be optimized with continuous, gradient-based optimal control in minutes, cutting energy use by about 8 percent and cost by 12.1 percent.

desk verdict A clean, fast continuous-optimal-control benchmark for district heating with open code, but the headline savings are internally consistent in an unvalidated dynamic model, so treat them as indicative, not measured. read the letter →

arxiv 2505.21679 v1 pith:VEL4VMHN submitted 2025-05-27 math.OC

Thomas Grandits , Stefano Coss , Gundolf Haase This is my paper

classification math.OC MSC 49J2049K2049M37

keywords districtheatingnetworksoptimalcontrolthermalplantschedulinglowtemperatureoperationdynamicenergypricinggradient-basedoptimizationPDE-constrainedstorage

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper argues that scheduling a district heating network can be treated as a smooth optimal control problem rather than as a mixed-integer program. The authors build a discretized heat-transport model, compress it into a differentiable solution operator that maps plant temperature settings to the temperature at every node, and then run a gradient-based optimizer that minimizes either total energy or operating cost while keeping every consumer above a contractual supply temperature. On a realistic Swiss alpine network with 150 substations and two plants, the optimizer lowers energy use by about 8 percent by running the network at the lowest temperatures that still satisfy consumers, and lowers operating cost by 12.1 percent when it shifts plant temperatures to follow hourly electricity prices. Each three-day schedule is computed in under five minutes on a desktop computer. What would follow if the paper is right is that district heating plant control can be re-optimized frequently, on real price signals, without integer variables or specialist solvers.

What carries the argument

The central object is the discrete solution operator for the network heat equation. After backward-Euler discretization, each time step requires solving the sparse linear system $\left(\frac{\rho c_p}{\Delta t} V + c_p G + S\right) y_t = \frac{\rho c_p}{\Delta t} y_{t-\Delta t} + S y_a$, where $V$ collects control volumes, $G$ encodes mass-flow convection, and $S$ encodes ambient heat loss through pipe insulation. The paper's move is to treat this linear solve as a differentiable operator, so the gradient of the loss with respect to every past plant temperature can be computed by backpropagation through all time steps, and the smooth problem is solved with a limited-memory quasi-Newton optimizer under an increasing penalty weight. The regularization $R_t(u) = \sum_t ((u(t)-u(t-\Delta t))/\Delta t)^2$ suppresses rapid changes in the plant controls, while the state constraints enter through the quadratic penalty $\max(0,c_i(u))^2$. This machinery is what replaces discrete on/off and integer decisions with a single continuous optimization.

What would settle it

Take the optimized plant-temperature schedules from the static and dynamic-pricing experiments and simulate them in a full dynamic network simulator that resolves time-varying mass flows (or run them on the real network); if the realized total energy and cost reductions are materially below 8% and 12.1%, or if consumer supply temperatures fall below $80\,^\circ\text{C}$ during load transients, the central claim is falsified.

Watch

Extended reading notes

Core claim

The central claim is that the physics of a district heating network can be folded into a differentiable mapping from plant control to network temperature, making the whole scheduling problem accessible to continuous optimization. The paper discretizes the convection-dominated heat equation on the network graph with a backward-Euler scheme, defines a solution operator $S$ that returns the temperature state for a given series of plant input temperatures, and minimizes the loss $L_e(y) = c_p \sum_i \int_T \dot m_i(t)(y_{p,s}^i - y_{p,r}^i) p(t, y_{p,s}^i, y_{p,r}^i)\, dt$ with $p=1$ for the energy case and the hourly electricity price, weighted by generation efficiencies $\alpha,\beta$, for the cost case. State constraints (consumer supply at least $80\,^\circ\text{C}$, consumer return at least $30\,^\circ\text{C}$, plants at most $140\,^\circ\text{C}$) are enforced by a continuation of quadratic penalties, and a first-order quadratic-variation term keeps the control smooth. With static pricing the optimizer drives plant temperatures down to the active constraint boundary and reports about 8 percent energy savings; with dynamic pricing it charges the network toward the temperature limit when electricity is cheap and lets it cool when prices are high, reporting a 12.1 percent cost reduction. The authors verify the solution operator in steady state against an established simulator with mean mismatch $0.23\,^\circ\text{C}$, and report a runtime below five minutes per three-day experiment.

Load-bearing premise

The load-bearing premise is that the dynamic temperature model (a backward-Euler discretization of the heat equation with constant mass flows taken from a single steady-state simulation) accurately represents what the real Swiss network does over the three-day horizon, because the optimized schedules and the 8% and 12.1% savings are never validated against a dynamic simulator or field data.

Editorial extensions

If this is right

Any district heating network that can be written as a directed graph with known pipe lengths and consumer demands fits the same solution-operator construction, so the method transfers beyond the tested alpine grid.
Because a three-day schedule costs under five minutes of desktop computation, the optimizer can be re-run on each updated day-ahead price profile, making price-aware operation an operational option rather than a planning exercise.
Under dynamic pricing the network itself acts as thermal storage: the optimizer deliberately stores heat during cheap hours near the $140\,^\circ\text{C}$ plant limit and depletes that stored energy before the horizon ends, which supports future integration of wind, solar, and power-to-heat.
The low-temperature scenario shows the main energy saving comes from holding every plant at the lowest temperature consistent with the $80\,^\circ\text{C}$ consumer requirement, implying the 8 percent figure is tied to how much margin the original schedule carried.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

A direct comparison with a mixed-integer formulation on the same network would quantify what the smooth approximation gives up by ignoring discrete plant, valve, and pump decisions.
The same differentiable solution-operator structure extends naturally to networks with explicit storage tanks, multiple heat sources, or power-to-heat units, since the control space remains just the time series of plant temperatures.
Because the optimizer depletes stored energy at the end of the fixed horizon, using this scheme as an online receding-horizon controller would require adding a terminal value or continuation term.
The most direct test of the 8% and 12.1% figures is to run the optimized schedules through a full dynamic network simulator and compare realized energy, cost, and temperature constraint satisfaction with the model's predictions.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

Colleague,

The short version: this is a clean, fast continuous-optimal-control pipeline for district heating, demonstrated on the open OpenDHN benchmark. The 8% and 12.1% savings are internally consistent reductions of the stated loss, but they are computed in a model that has only been checked against PyDHN in steady state. Read the percentages as indicative, not as measured performance.

What is actually new is the differentiable solver. The authors build a sparse linear solution operator for the discretized convection-diffusion network model, implement it in PyTorch with automatic differentiation, and publish the code. That makes full-horizon optimal control with L-BFGS feasible in minutes on a desktop, which is a practical step beyond the usual MILP treatments. The model itself is borrowed from Krug et al., and continuous DHN optimization appears in Bavière and Vallée, so the novelty is in the implementation, the open benchmark, and the dynamic-pricing storage study.

The paper does several things well. It is transparent about the model heritage, openly states that mass flows are taken from a single steady-state PyDHN simulation and used for the whole horizon, and even notes a substation modification that was needed to stabilize the benchmark. The steady-state temperature mismatch against PyDHN is 0.23 °C mean, which shows the solution operator is implemented correctly. The code and data are linked, and the citation pattern is reasonable. That is genuine, reproducible work.

The soft spots are real, and they are mainly about what the savings claim can support. The baseline 'Init y' schedule is never precisely specified, so the 8% reduction is relative to an unknown object. More importantly, the dynamic temperature model holds mass flows constant while consumer demand curves vary over three days; in a real network, changing withdrawals change flow setpoints, transport delays, and pipe losses. The dynamic-pricing experiment deliberately charges and discharges thermal storage in the pipes, so fixed flows can distort the 12.1% figure. The authors validate their surrogate only at steady state, and the optimized trajectories are never checked against a dynamic simulator or field data. None of this makes the method wrong; it makes the headline numbers scenario-illustrative rather than validated.

Who is this for? Anyone working on DHN scheduling, and anyone who wants a worked example of differentiable simulation for PDE-constrained optimal control. It deserves a serious referee, with the main asks being: specify the baseline, add a dynamic validation against PyDHN, and release the full reproduction scripts with a pinned environment.

I'd take it as a benchmark and engineering contribution, not as a claim about actual savings in a real network.

Referee Report

4 major / 5 minor

Summary. The paper formulates the control of district heating network (DHN) plant supply temperatures as a continuous optimal control problem. The authors define a linearized thermodynamic network model (Eq. 3), construct a solution operator that maps plant temperature controls to network temperatures, and minimize an energy or dynamic-pricing cost (Eq. 5) using L-BFGS with penalty methods for constraints. The method is applied to the OpenDHN Verbier benchmark with synthetic consumer demand and Swiss day-ahead electricity prices. The paper reports an 8% reduction in energy consumption for static pricing and a 12.1% reduction in operating cost under dynamic pricing, with total runtime under 5 minutes per three-day experiment on a desktop computer, and interprets these results as evidence that continuous optimization is a practical alternative to mixed-integer programming for DHN scheduling.

Significance. If the reported results are correct, the paper demonstrates an attractive computational approach: a sparse linear solution operator differentiated through PyTorch, with a runtime of minutes, is a genuine practical contribution. The authors also ship open-source code, and the steady-state validation against PyDHN (mean mismatch 0.23 °C, Fig. 4) is a useful sanity check. However, the central quantitative claims of 8% and 12.1% savings rest on a dynamic model whose fidelity is not established in time-dependent settings, and the savings are measured relative to an unspecified baseline schedule. The paper is an honest and clearly written application study, but its headline conclusions currently outrun the evidence presented.

major comments (4)

[§3.1, §3.2, Eq. (3), Fig. 4] The dynamic temperature model is not validated in time-dependent conditions. Mass flows are computed once from a steady-state PyDHN simulation and held fixed while consumer heat demands vary over the three-day horizon, yet the solution operator S is checked against PyDHN only in steady state (Fig. 4, mean absolute mismatch 0.23 °C). Because the 8% and 12.1% savings are computed entirely with this model, an error in the dynamic temperature field would propagate directly into the headline figures. The authors should validate the optimized trajectories against a dynamic simulator (e.g., time-dependent PyDHN) or field measurements, or explicitly restrict the claims to the validity of the constant-mass-flow model.
[§3.2, Figs. 5–9] The savings figures are relative to an unspecified initial schedule. The paper repeatedly compares 'Init y' with 'Final y' but never defines how the initial plant temperature schedule was constructed. Without this baseline, the 8% and 12.1% numbers are not reproducible, and the reader cannot judge whether they reflect a realistic operating point or an artificially poor one. Please specify the baseline schedule (e.g., constant temperatures, a heuristic rule, or the OpenDHN benchmark schedule) and report the absolute energy and cost values for both baseline and optimized cases.
[§2.3, Eq. (5), §3.2] The reported savings are reductions of the objective function (5) that is minimized, not measured savings from an independent evaluation. This is not a flaw in the optimization methodology, but it means the abstract's wording 'demonstrate ... savings' overstates the evidence: the optimized solution is evaluated with the same loss that defines optimality. The paper should state explicitly that the 8% and 12.1% figures are model-objective reductions, and should provide an out-of-sample or simulator-based evaluation if the intention is to claim physical energy or cost savings.
[§3.1, Eq. (6)] The values of the efficiencies α and β used in the dynamic-pricing experiment are not reported. The cost saving of 12.1% depends directly on p(t,y_i,y_o) through Eq. (6), and the paper does not state whether β = 0 (no heat-to-power recovery) or a positive value was used. Please report the exact values of α and β, and ideally a sensitivity analysis, so that the dynamic-pricing result is reproducible and interpretable.

minor comments (5)

[Abstract / Introduction] The sentence 'these has been a central role in improving energy efficiency' should be rephrased, e.g., 'they have played a central role'.
[§2.2, Eq. (3)] In Eq. (3), the right-hand side 'cρ y_{t-Δt}' should read 'ρ c_p y_{t-Δt}' (or 'c_p ρ'), matching the definition of c_p as specific heat capacity in Table 1.
[§2.3, Eq. (9)] The penalty term is written as ṁ(c_i(u)), which clashes with the mass-flow notation ṁ used throughout the paper. Using a different symbol, such as g(c_i(u)) or p(c_i(u)), would avoid confusion.
[§3.1, Eq. (11)] The sentence introducing the constraints is grammatically incomplete: 'In short, the following the consumer input and output temperatures have to be fulfilled' should be 'In short, the following consumer input and output temperatures must be fulfilled'.
[Fig. 4] The histogram would benefit from a y-axis label (e.g., 'Count') and a caption stating the number of nodes compared.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the savings figures are the optimized objective values of a model-based benchmark, not predictions derived from their own inputs.

full rationale

The derivation chain is: the thermodynamic state operator S is defined by the discretized model (Eq. 3), the loss Le is defined in Eq. (5) as the energy/cost of operation, and the paper solves min_u Le(S(u)) subject to constraints and regularization. The reported savings (8% and 12.1%) are the relative reductions of Le between the initial operating point and the optimized control. This is the objective value of the optimization itself, not an externally predicted quantity fitted to the loss, and it is the standard structure of an optimal-control benchmark. No parameter of the model or loss is fitted to the reported savings, so there is no fitted-input-called-prediction step. The solution operator is independently checked against the PyDHN simulator in steady state (Fig. 4, mean mismatch 0.23 °C), giving an external anchor for the model; the dynamic temperature field is not separately validated, which is a correctness/validation limitation rather than a circularity. There are no load-bearing self-citations, no imported uniqueness theorems, and no ansatz smuggled in via citation. The paper is openly a model-based optimization study whose claims are conditional on the thermodynamic model's fidelity.

Assumptions & free parameters 5 free parameters · 5 assumptions · 0 invented entities

The central claims rest on the chosen temperature model, the constant mass-flow assumption, an undefined baseline schedule, and a set of unstated efficiency and regularization parameters. No new physical entities are introduced.

free parameters (5)

Baseline plant temperature schedule ('Init y') = not specified
The reported 8% and 12.1% savings are reductions relative to this initial control; its definition is not given, so the headline numbers depend on it.
Power-to-heat efficiency α and heat-to-power efficiency β = not stated
Eq. (6) scales the dynamic price with α and β, but the values used in the dynamic pricing experiment are not reported; only α, β > 0 is assumed.
Substation S132 mass flow modification = 0.03 kg/s
Manually changed from near zero because it caused major deviations; the effect on the optimization results is not quantified.
Regularization weight for Rt(u) = not stated
The Tikhonov term (Eq. 7) is introduced but never referenced in the experiments, so it is unclear if and how it is applied.
Synthetic demand profile noise parameters = not fully specified
Multiplicative Fourier-domain noise creates 150 consumer curves; the amplitude distribution is not described, so the demand data is not exactly reproducible.

assumptions (5)

domain assumption Convection-dominated flow with negligible diffusion and backward Euler time stepping (Eq. 3) adequately models DHN temperature dynamics.
The optimization uses this discretized model as the physical truth; only steady-state validation against PyDHN is provided, with no dynamic validation.
domain assumption Mass flows are constant and taken from a single steady-state PyDHN simulation.
Sec. 3.1 states 'This steady mass flow is used throughout this work'; temperature-dependent flow changes are ignored.
standard math Penalty method with increasing λp converges to the constrained optimum (Boyd and Vandenberghe).
Eqs. (9)-(10) rely on the classic penalty method equivalence as λp tends to infinity.
domain assumption The OpenDHN anonymized network and synthetic consumer profiles are representative of real operation.
The network has 150 instead of 165 substations and the demand curves are synthetic variations of a Belgian profile, so the setup is an approximation.
domain assumption EPEX day-ahead prices linearly interpolated to 15-minute intervals represent dynamic energy tariffs.
Sec. 3.1 uses one three-day price series; no sensitivity to the price scenario is given.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Optimal dynamic thermal plant control: A study and benchmark." pith.science (2026). https://pith.science/paper/VEL4VMHN

@misc{pith2026250521679,
  author       = {Pith},
  title        = {Pith review of: Optimal dynamic thermal plant control: A study and benchmark},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/VEL4VMHN}},
  note         = {Machine review of arXiv:2505.21679}
}

read the original abstract

District heating networks play a vital role in thermal energy supply in many countries. Thus, it comes to no surprise that these has been a central role in improving energy efficiency for private and public energy suppliers alike around the globe. Many studies have previously investigated the potential of energy saving by low temperature operation of the DHN and the integration of renewable energies. Many other studies consider this problem in terms of mixed integer lin-ear programming. Here, we instead investigate the utilization of well-established continuous optimization methods to improve DHN operation efficiency. We demonstrate that optimal control is able to model low temperature operation of a DHN for savings of around 8%, but can even further improve its operation when considering dynamic energy pricing, reducing the cost of operation by roughly 12%. We demonstrate the applicability of this method in a realistic, openly available network in Switzerland (OpenDHN), with a total runtime of less than 5 minutes on a standard desktop com-puter per experiment.

Figures

Figures reproduced from arXiv: 2505.21679 by the authors.

**Figure 1.** Illustration of a DHN graph, showing the duplicated graph structure and separation of nodes and edges. The nodes are separated into supply (black) and return nodes (blue), while the edges can either be considered supply (black), return (blue), consumer (purple) and producer (red) edges. Figure reproduced from [8]. a node Vi [12], we write the discretized form as ( ρcpViy˙i + cp P j m˙ jyj = kl(yi − ya) + ϕ i /∈ BC y… view at source ↗

**Figure 2.** OpenDHN Net created from data of Verbier in Switzerland [18]. The pipes of the DHN are shown as lines, the blue dots represent consumer sites and the red stars visualize the thermal supply plants. The color of the lines represents the mass flow through the network (logarithmic scale). losses in Sec. 2.3. The automatic differentiation of the sparse solution operator in (3) was implemented and published on github for… view at source ↗

**Figure 3.** We show the average consumer model unfiltered in blue [20] and its low-pass filtered version in orange (top), along with the created individual variations of the single substations as transparent black lines (bottom), over the chosen three day period. 3.2 Results −4 −2 0 2 4 6 Temperature mismatch [°C] 0 100 200 300 Count [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗

Figures from the paper (4 more)

**Figure 4.** Figure 4: Histogram of the temperature mismatch (x-axis) of our solution operator S against the PyDHN simulation on the OpenDHN benchmark. We first verified our thermal computations by testing the solution operator in steady state (y˙ = 0) by applying the temperature deltas of e…

**Figure 5.** Figure 5: Optimal control result of the analyzed problem of Verbier over the chosen three day period. The top plot shows the input temperature of the network at the two thermal plants in different colors before (dashed) and after (solid) optimization. The middle plot shows the m…

**Figure 6.** Figure 6: Same as [PITH_FULL_IMAGE:figures/full_fig_p005_6.png]

**Figure 9.** Figure 9: Same as [PITH_FULL_IMAGE:figures/full_fig_p006_9.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

25 extracted references · 22 canonical work pages

[1]

The role of district heating in future renewable energy systems

H. Lund, B. Möller, B. V. Mathiesen, and A. Dyrelund. “The role of district heating in future renewable energy systems”. In: Energy 35.3 (Mar. 2010), pp. 1381–1390.issn: 0360-5442. doi: 10.1016/j.energy.2009.11.023

doi:10.1016/j.energy.2009.11.023 2010
[2]

Energy Storage and Smart Energy Systems

Henrik Lund, Poul Alberg Østergaard, David Connolly, Iva Ridjan, Brian Vad Mathiesen, Frede Hvelplund, Jakob Zinck Thellufsen, and Peter Sorknæs. “Energy Storage and Smart Energy Systems”. en. In:International Journal of Sustainable Energy Planning and Management11 (Oct. 2016), pp. 3–14. issn: 2246-2929. doi: 10.5278/ijsepm.2016.11.2

work page doi:10.5278/ijsepm.2016.11.2 2016
[3]

The benefits of 4th generation district heating in a 100% renewable energy system

Peter Sorknæs, Poul Alberg Østergaard, Jakob Zinck Thel- lufsen, Henrik Lund, Steffen Nielsen, Søren Djørup, and Karl Sperling. “The benefits of 4th generation district heating in a 100% renewable energy system”. In:Energy 213 (Dec. 2020), p. 119030. issn: 0360-5442. doi: 10 . 1016 / j . energy . 2020 . 119030

work page 2020
[4]

The potential of power-to-heat in Swedish district heating systems

Gerald Schweiger, Jonatan Rantzer, Karin Ericsson, and Patrick Lauenburg. “The potential of power-to-heat in Swedish district heating systems”. In:Energy 137 (Oct. 2017), pp. 661–

work page 2017
[5]

District heating networks – dynamic simulation and optimal operation

Jan Mohring, Dominik Linn, Matthias Eimer, Markus Rein, and Norbert Siedow. “District heating networks – dynamic simulation and optimal operation”. In:Mathematical Model- ing, Simulation and Optimization for Power Engineering and Management (2021), pp. 303–325

work page 2021
[6]

Optimal operation of large district heating networks through fast fluid-dynamic simulation

Elisa Guelpa, Claudia Toro, Adriano Sciacovelli, Roberto Melli, Enrico Sciubba, and Vittorio Verda. “Optimal operation of large district heating networks through fast fluid-dynamic simulation”. In:Energy 102 (May 2016), pp. 586–595.issn: 0360-5442. doi: 10.1016/j.energy.2016.02.058

work page doi:10.1016/j.energy.2016.02.058 2016
[7]

Optimal Temperature Control of Large Scale District Heating Networks

Roland Bavière and Mathieu Vallée. “Optimal Temperature Control of Large Scale District Heating Networks”. In:Energy Procedia. 16th International Symposium on District Heating and Cooling, DHC2018 149 (Sept. 2018), pp. 69–78. issn: 1876-6102. doi: 10.1016/j.egypro.2018.08.170

work page doi:10.1016/j.egypro.2018.08.170 2018
[8]

Non- linear optimization of district heating networks

Richard Krug, Volker Mehrmann, and Martin Schmidt. “Non- linear optimization of district heating networks”. en. In:Opti- mization and Engineering22.2 (June 2021), pp. 783–819.issn: 1573-2924. doi: 10.1007/s11081-020-09549-0

work page doi:10.1007/s11081-020-09549-0 2021

Show all 25 references

[9]

Springer, 2021

Andrea Manzoni, Alfio Quarteroni, and Sandro Salsa.Optimal control of partial differential equations. Springer, 2021

2021
[10]

Graph theory

John Adrian Bondy and Uppaluri Siva Ramachandra Murty. Graph theory. Springer Publishing Company, Incorporated, 2008

2008
[11]

A state of art review on the district heating systems

Abdur Rehman Mazhar, Shuli Liu, and Ashish Shukla. “A state of art review on the district heating systems”. In:Renew- able and Sustainable Energy Reviews96 (Nov. 2018), pp. 420–

2018
[12]

Thermo-fluid dynamic model of large district heating net- works for the analysis of primary energy savings

Elisa Guelpa, Adriano Sciacovelli, and Vittorio Verda. “Thermo-fluid dynamic model of large district heating net- works for the analysis of primary energy savings”. en. In: Energy. Shaping research in gas-, heat- and electric- energy infrastructures 184 (Oct. 2019), pp. 34–44....

2019 doi
[13]

Fredi Tröltzsch.Optimal control of partial differential equa- tions: theory, methods, and applications. Vol. 112. American Mathematical Soc., 2010

2010
[14]

Modern regularization methods for inverse problems

Martin Benning and Martin Burger. “Modern regularization methods for inverse problems”. en. In:Acta Numerica27 (May 2018), pp. 1–111.issn: 0962-4929, 1474-0508.doi: 10.1017/ S0962492918000016

2018
[15]

Cambridge university press, 2004.isbn: 978-0-521-83378-3

Stephen Boyd and Lieven Vandenberghe.Convex optimization. Cambridge university press, 2004.isbn: 978-0-521-83378-3

2004
[16]

PyTorch: An Im- perative Style, High-Performance Deep Learning Library

Adam Paszke, Sam Gross, Francisco Massa, Adam Lerer, James Bradbury, Gregory Chanan, Trevor Killeen, Zeming Lin, Natalia Gimelshein, Luca Antiga, Alban Desmaison, An- dreas Kopf, Edward Yang, Zachary DeVito, Martin Raison, Alykhan Tejani, Sasank Chilamkurthy, Benoit Steiner, L...

2019
[17]

A Limited Memory Algorithm for Bound Constrained Optimization

R. Byrd, P. Lu, J. Nocedal, and C. Zhu. “A Limited Memory Algorithm for Bound Constrained Optimization”. In:SIAM Journal on Scientific Computing16.5 (Sept. 1995), pp. 1190–

1995
[18]

Roberto Boghetti and Jérôme Henri Kämpf.OpenDHN data. eng. June 2023.doi: 10.5281/zenodo.8058601

2023 doi
[19]

Verification of an open-source Python library for the simulation of district heating networks with complex topologies

Roberto Boghetti and Jérôme H. Kämpf. “Verification of an open-source Python library for the simulation of district heating networks with complex topologies”. In:Energy 290 6 (2024), p. 130169.issn: 0360-5442. doi: https://doi.org/10. 1016/j.energy.2023.130169

2024
[20]

Chaïm De Mulder.2021 electricity and heat demand data for a city district (Belgium). nld. Jan. 2022.doi: 10.5281/zenodo. 5155659

2021 doi
[21]

org / data / european - wholesale - electricity-price-data

Ember Energy.European Wholesale Electricity Price Data, https : / / ember - energy . org / data / european - wholesale - electricity-price-data
[22]

Perspectives for low- temperature waste heat recovery

Z. Y. Xu, R. Z. Wang, and Chun Yang. “Perspectives for low- temperature waste heat recovery”. In:Energy 176 (June 2019), pp. 1037–1043. issn: 0360-5442. doi: 10.1016/j.energy.2019. 04.001. 7

2019 doi
[439]

doi: 10.1016/j.rser.2018.08.005

issn: 1364-0321. doi: 10.1016/j.rser.2018.08.005

2018 doi
[669]

doi: 10.1016/j.energy.2017.02.075

issn: 0360-5442. doi: 10.1016/j.energy.2017.02.075

2017 doi
[1208]

doi: 10.1137/0916069

issn: 1064-8275. doi: 10.1137/0916069

doi

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.

[1] [1]

The role of district heating in future renewable energy systems

H. Lund, B. Möller, B. V. Mathiesen, and A. Dyrelund. “The role of district heating in future renewable energy systems”. In: Energy 35.3 (Mar. 2010), pp. 1381–1390.issn: 0360-5442. doi: 10.1016/j.energy.2009.11.023

doi:10.1016/j.energy.2009.11.023 2010

[2] [2]

Energy Storage and Smart Energy Systems

Henrik Lund, Poul Alberg Østergaard, David Connolly, Iva Ridjan, Brian Vad Mathiesen, Frede Hvelplund, Jakob Zinck Thellufsen, and Peter Sorknæs. “Energy Storage and Smart Energy Systems”. en. In:International Journal of Sustainable Energy Planning and Management11 (Oct. 2016), pp. 3–14. issn: 2246-2929. doi: 10.5278/ijsepm.2016.11.2

work page doi:10.5278/ijsepm.2016.11.2 2016

[3] [3]

The benefits of 4th generation district heating in a 100% renewable energy system

Peter Sorknæs, Poul Alberg Østergaard, Jakob Zinck Thel- lufsen, Henrik Lund, Steffen Nielsen, Søren Djørup, and Karl Sperling. “The benefits of 4th generation district heating in a 100% renewable energy system”. In:Energy 213 (Dec. 2020), p. 119030. issn: 0360-5442. doi: 10 . 1016 / j . energy . 2020 . 119030

work page 2020

[4] [4]

The potential of power-to-heat in Swedish district heating systems

Gerald Schweiger, Jonatan Rantzer, Karin Ericsson, and Patrick Lauenburg. “The potential of power-to-heat in Swedish district heating systems”. In:Energy 137 (Oct. 2017), pp. 661–

work page 2017

[5] [5]

District heating networks – dynamic simulation and optimal operation

Jan Mohring, Dominik Linn, Matthias Eimer, Markus Rein, and Norbert Siedow. “District heating networks – dynamic simulation and optimal operation”. In:Mathematical Model- ing, Simulation and Optimization for Power Engineering and Management (2021), pp. 303–325

work page 2021

[6] [6]

Optimal operation of large district heating networks through fast fluid-dynamic simulation

Elisa Guelpa, Claudia Toro, Adriano Sciacovelli, Roberto Melli, Enrico Sciubba, and Vittorio Verda. “Optimal operation of large district heating networks through fast fluid-dynamic simulation”. In:Energy 102 (May 2016), pp. 586–595.issn: 0360-5442. doi: 10.1016/j.energy.2016.02.058

work page doi:10.1016/j.energy.2016.02.058 2016

[7] [7]

Optimal Temperature Control of Large Scale District Heating Networks

Roland Bavière and Mathieu Vallée. “Optimal Temperature Control of Large Scale District Heating Networks”. In:Energy Procedia. 16th International Symposium on District Heating and Cooling, DHC2018 149 (Sept. 2018), pp. 69–78. issn: 1876-6102. doi: 10.1016/j.egypro.2018.08.170

work page doi:10.1016/j.egypro.2018.08.170 2018

[8] [8]

Non- linear optimization of district heating networks

Richard Krug, Volker Mehrmann, and Martin Schmidt. “Non- linear optimization of district heating networks”. en. In:Opti- mization and Engineering22.2 (June 2021), pp. 783–819.issn: 1573-2924. doi: 10.1007/s11081-020-09549-0

work page doi:10.1007/s11081-020-09549-0 2021