pith. sign in

arxiv: 2501.10859 · v2 · submitted 2025-01-18 · 📡 eess.SY · cs.LG· cs.SY· math.OC

What price to pay? Auto-tuning a building MPC controller for optimal economic cost

Jiarui Yu , Jicheng Shi , Wenjie Xu , Colin N. Jones This is my paper

Pith reviewed 2026-05-23 05:13 UTC · model grok-4.3

classification 📡 eess.SY cs.LGcs.SYmath.OC
keywords Model Predictive ControlBayesian OptimizationDemand Side ManagementBuilding Energy ControlEconomic OptimizationHyperparameter TuningElectricity Cost Reduction
0
0 comments X

The pith

Constrained Bayesian Optimization auto-tunes building MPC controllers to cut electricity costs by 26.9 percent versus rule-based methods.

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

The paper shows that Constrained Bayesian Optimization can automatically select the hyperparameters of a Model Predictive Controller so that the controller minimizes electricity bills under complex demand-side management pricing. In the reported case study the resulting controller achieved 26.90 percent lower costs than a simple rule-based strategy and 17.46 percent lower costs than an MPC that had been tuned by hand. The same optimization framework also identified DSM contract choices that cut monthly bills by as much as 20.18 percent. The central practical result is therefore a data-driven route from measured building behavior and pricing data to lower consumer electricity expenditure.

Core claim

Constrained Bayesian Optimization applied to the hyperparameter tuning of a building Model Predictive Controller produces an optimized controller whose closed-loop electricity cost is 26.90 percent lower than that of a rule-based baseline and 17.46 percent lower than that of a manually tuned MPC; when the same method is used to select among real DSM contracts the monthly bill can be reduced by up to 20.18 percent.

What carries the argument

Constrained Bayesian Optimization (CONFIG) that treats MPC hyperparameter selection as a black-box optimization problem whose objective is measured economic cost.

If this is right

  • MPC controllers whose hyperparameters are chosen by CONFIG deliver lower electricity expenditure than both rule-based logic and manually tuned MPC under the same pricing.
  • Selecting the best DSM program with the same optimization method produces additional bill reductions of up to 20.18 percent per month.
  • The economic benefit is obtained without requiring the designer to perform manual trial-and-error tuning of the controller.
  • The approach is demonstrated on a real building and real utility contracts, indicating that the method can be applied to actual consumer installations.

Where Pith is reading between the lines

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

  • The same CONFIG loop could be rerun periodically as new pricing contracts or building usage patterns appear, turning the tuning step into an ongoing service rather than a one-time design task.
  • If the underlying thermal model is updated from streaming sensor data, the economic savings might be maintained even when the building or its occupancy changes.
  • Extending the method to multi-building or district-scale problems would require only that the cost function be defined at the aggregate level; the optimization machinery itself would remain unchanged.

Load-bearing premise

The building thermal model used inside the MPC must accurately capture the real thermal dynamics and disturbances under the tested pricing signals.

What would settle it

Deploy the CONFIG-tuned MPC on the physical building for one or more billing periods under the same pricing signals and compare the measured electricity cost against the rule-based controller; a realized saving materially below 26.9 percent would falsify the reported performance claim.

Figures

Figures reproduced from arXiv: 2501.10859 by Colin N. Jones, Jiarui Yu, Jicheng Shi, Wenjie Xu.

Figure 1
Figure 1. Figure 1: An overview of the approach for developing the performance-oriented MPC controller tuning in building control [PITH_FULL_IMAGE:figures/full_fig_p006_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: An overview of the CONFIG Bayesian Optimization method. The [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Temperature constraints illustration 0.0 0.2 0.4 0.6 0.8 1.0 Modulated heat pump power 0 500 1000 1500 2000 2500 3000 3500 Electricity power (W) Modulated input and actual electricity power in January Bang-bang control MPC control Mask MPC [PITH_FULL_IMAGE:figures/full_fig_p010_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Start up cost and concept of the Mask MPC [PITH_FULL_IMAGE:figures/full_fig_p010_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Cumulative distribution function constraint of PMV [PITH_FULL_IMAGE:figures/full_fig_p012_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Electricity cost comparison of three different controllers 16 18 20 22 24 26 T e m p e r a t u r e ( C) Performances of different controllers in January (section, Baseline:345 , QP MPC:304.25 , Mask MPC:269.83 , MIQP MPC:273.54 ) Indoor Temperature (Baseline) Indoor Temperature (Mask MPC) Indoor Temperature (QP MPC) Indoor Temperature (MIQP MPC) Constraint 0.0 0.5 1.0 Modulated heat pump power Mask MPC (u … view at source ↗
Figure 7
Figure 7. Figure 7: Performance comparison of three different controllers the customers in Belgium to better choose a electricity contract, i.e. a contract that costs the least. After tuning the Mask MPC controller for each contract and each month, as shown in Table.5, it is evident that choosing a proper electricity bill is vital as it could help customers save up to 53.70 euros a month and that the best bills are reduced up… view at source ↗
Figure 8
Figure 8. Figure 8: Cumulative distributive function of absolute PMV during occupied [PITH_FULL_IMAGE:figures/full_fig_p015_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Electricity costs for two typical months with di [PITH_FULL_IMAGE:figures/full_fig_p016_9.png] view at source ↗
read the original abstract

Demand-side management (DSM) programs introduce complex pricing, requiring advanced control for cost minimization. Model Predictive Control (MPC) offers a solution but its performance hinges on appropriate hyperparameter tuning. We propose using Constrained Bayesian Optimization (CONFIG) to automate this process. In a case study, our optimized MPC reduced electricity costs by 26.90% compared to a rule-based controller and by 17.46% versus an manually tuned MPC. Analysis of real contracts further showed that optimal DSM program selection can lower monthly bills by up to 20.18%, demonstrating a data-driven path to significant consumer savings.

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 / 1 minor

Summary. The manuscript proposes using Constrained Bayesian Optimization (CONFIG) to automatically tune the hyperparameters of a building MPC controller aimed at minimizing electricity costs under complex demand-side management (DSM) pricing. In a case study, the optimized MPC is reported to reduce costs by 26.90% versus a rule-based controller and 17.46% versus a manually tuned MPC; separate analysis of real contracts indicates that optimal DSM program selection can lower monthly bills by up to 20.18%.

Significance. If the simulation results hold under real conditions, the work offers a concrete, automated method for achieving measurable economic savings in building energy systems by replacing manual MPC tuning and by selecting among DSM contracts. The empirical case-study numbers provide a clear benchmark for the value of CONFIG over baseline approaches.

major comments (2)
  1. [Abstract] Abstract and case-study results: the headline savings (26.90% and 17.46%) are presented without any reported model validation, error bars, number of trials, or statistical significance; these omissions directly affect the ability to assess whether the claimed cost reductions support the central contribution.
  2. [Case study (simulation)] Case-study simulation setup: both the MPC predictor and the plant use the identical building thermal model, so the reported economic performance is obtained under perfect model match; any deviation in heat-transfer coefficients, occupancy, or HVAC nonlinearities under the tested price signals would invalidate the savings figures.
minor comments (1)
  1. [Abstract] Abstract contains a grammatical error: 'versus an manually tuned MPC' should read 'versus a manually tuned MPC'.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments, which highlight important aspects of result presentation and simulation assumptions. We address each major comment below and indicate where revisions will be made.

read point-by-point responses

  1. Referee: [Abstract] Abstract and case-study results: the headline savings (26.90% and 17.46%) are presented without any reported model validation, error bars, number of trials, or statistical significance; these omissions directly affect the ability to assess whether the claimed cost reductions support the central contribution.

    Authors: We agree that the abstract would benefit from additional qualifiers on the nature of the results. The reported savings come from a deterministic simulation study whose setup (including the single optimization run and perfect model match) is detailed in Sections 4 and 5. In the revision we will expand the abstract to state that the figures are simulation-based and to reference the case-study section for validation details. Because the CONFIG procedure was executed once per scenario, we cannot retroactively add error bars or statistical significance tests; we will instead clarify the number of function evaluations performed during optimization. revision: partial

  2. Referee: [Case study (simulation)] Case-study simulation setup: both the MPC predictor and the plant use the identical building thermal model, so the reported economic performance is obtained under perfect model match; any deviation in heat-transfer coefficients, occupancy, or HVAC nonlinearities under the tested price signals would invalidate the savings figures.

    Authors: This observation is correct. The case study deliberately employs an identical linear thermal model for both the controller and the plant to isolate the benefit of hyperparameter tuning. We will add an explicit limitations paragraph in the discussion section that states the perfect-match assumption, notes that real deviations in heat-transfer coefficients, occupancy, or actuator nonlinearities could reduce the observed savings, and outlines future work on robust or adaptive MPC formulations. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical case-study comparisons only

full rationale

The manuscript reports simulation-based cost reductions from an optimized MPC versus rule-based and manual baselines, plus a DSM-contract ranking. No derivation chain, first-principles equations, or predictions appear; results are direct empirical outputs of the CONFIG tuning procedure evaluated in the same model. No self-citation load-bearing steps, fitted inputs renamed as predictions, or ansatz smuggling are present. The central claims therefore remain independent of the paper's own inputs by construction.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Based on abstract only; no explicit free parameters, axioms, or invented entities are described beyond standard assumptions of MPC and Bayesian optimization.

pith-pipeline@v0.9.0 · 5644 in / 999 out tokens · 66980 ms · 2026-05-23T05:13:56.698042+00:00 · methodology

discussion (0)

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

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Bayesian Optimization with Structured Measurements: A Vector-Valued RKHS Framework

    cs.LG 2026-05 unverdicted novelty 7.0

    Proposes a vector-valued RKHS framework for Bayesian optimization with structured measurements, deriving concentration bounds and UCB-based regret guarantees that recover sublinear rates.

Reference graph

Works this paper leans on

41 extracted references · 41 canonical work pages · cited by 1 Pith paper

  1. [1]

    U. N. E. Programme, 2022 global status report for buildings and con- struction: Towards a zero-emission, efficient and resilient buildings and construction sector (2022-11)

    work page 2022

  2. [2]

    Davies, Benchmarking commercial building energy use per square foot, Altenergymag.com (Jun

    M. Davies, Benchmarking commercial building energy use per square foot, Altenergymag.com (Jun. 8, 2019)

    work page 2019

  3. [3]

    Z. Liu, X. Zhang, Y . Sun, Y . Zhou, Advanced controls on energy reliability, flexibility and occupant-centric control for smart and energy-e fficient buildings, Energy and Buildings 297 (2023) 113436. doi:https://doi. org/10.1016/j.enbuild.2023.113436

    work page doi:10.1016/j.enbuild.2023.113436 2023

  4. [4]

    C. Eid, E. Koliou, M. Valles, J. Reneses, R. Hakvoort, Time-based pricing and electricity demand response: Existing barriers and next steps, Utili- ties Policy 40 (2016) 15–25. doi:https://doi.org/10.1016/j.jup. 2016.04.001

    work page doi:10.1016/j.jup 2016

  5. [5]

    S. Yang, H. Oliver Gao, F. You, Model predictive control for demand- and market-responsive building energy management by leveraging active latent heat storage, Applied Energy 327 (2022) 120054. doi:https: //doi.org/10.1016/j.apenergy.2022.120054

    work page doi:10.1016/j.apenergy.2022.120054 2022

  6. [6]

    S. Yang, M. P. Wan, W. Chen, B. F. Ng, S. Dubey, Model predictive control with adaptive machine-learning-based model for building energy efficiency and comfort optimization, Applied Energy 271 (2020) 115147. doi:https://doi.org/10.1016/j.apenergy.2020.115147. 13 /uni00000031/uni00000052/uni00000059/uni00000027/uni00000048/uni00000046/uni0000002d/uni000000...

    work page doi:10.1016/j.apenergy.2020.115147 2020

  7. [7]

    Cigler, S

    J. Cigler, S. Pr ´ıvara, Z. V´aˇna, E. ˇZ´aˇcekov´a, L. Ferkl, Optimization of predicted mean vote index within model predictive control framework: Computationally tractable solution, Energy and Buildings 52 (2012) 39–49. doi:https://doi.org/10.1016/j.enbuild.2012.05.022

    work page doi:10.1016/j.enbuild.2012.05.022 2012

  8. [9]

    Oldewurtel, A

    F. Oldewurtel, A. Ulbig, A. Parisio, G. Andersson, M. Morari, Reducing peak electricity demand in building climate control using real-time pricing and model predictive control, in: 49th IEEE Conference on Decision and Control (CDC), 2010, pp. 1927–1932. doi:10.1109/CDC.2010. 5717458

    work page doi:10.1109/cdc.2010 2010

  9. [10]

    N. T. Mbungu, R. M. Naidoo, R. Bansal, Real-time electricity pricing: Tou-mpc based energy management for commercial buildings, Energy Procedia 105 (2017) 3419–3424, 8th International Conference on Applied Energy, ICAE2016, 8-11 October 2016, Beijing, China. doi:https: //doi.org/10.1016/j.egypro.2017.03.781

    work page doi:10.1016/j.egypro.2017.03.781 2017

  10. [11]

    Y . Pan, M. Zhu, Y . Lv, Y . Yang, Y . Liang, R. Yin, Y . Yang, X. Jia, X. Wang, F. Zeng, S. Huang, D. Hou, L. Xu, R. Yin, X. Yuan, Building energy simu- lation and its application for building performance optimization: A review of methods, tools, and case studies, Advances in Applied Energy 10 (2023) 100135. doi:https://doi.org/10.1016/j.adapen.2023.100135

    work page doi:10.1016/j.adapen.2023.100135 2023

  11. [12]

    Z. Wang, M. Broccardo, J. Song, Hamiltonian monte carlo methods for subset simulation in reliability analysis, Structural Safety 76 (2019) 51–67. doi:https://doi.org/10.1016/j.strusafe.2018.05.005

    work page doi:10.1016/j.strusafe.2018.05.005 2019

  12. [13]

    Y . Lian, J. Shi, M. Koch, C. N. Jones, Adaptive robust data-driven building control via bilevel reformulation: An experimental result, IEEE Transactions on Control Systems Technology 31 (6) (2023) 2420–2436. doi:10.1109/TCST.2023.3259641

    work page doi:10.1109/tcst.2023.3259641 2023

  13. [14]

    Massa Gray, M

    F. Massa Gray, M. Schmidt, Thermal building modelling using gaussian processes, Energy and Buildings 119 (2016) 119–128. doi:https:// doi.org/10.1016/j.enbuild.2016.02.004

    work page doi:10.1016/j.enbuild.2016.02.004 2016

  14. [15]

    C. Fan, J. Wang, W. Gang, S. Li, Assessment of deep recurrent neural network-based strategies for short-term building energy predictions, Ap- plied Energy 236 (2019) 700–710. doi:https://doi.org/10.1016/ j.apenergy.2018.12.004

    work page 2019

  15. [17]

    S. Yang, H. O. Gao, F. You, Integrated optimization in operations control and systems design for carbon emission reduction in building electrification with distributed energy resources, Advances in Applied Energy (2023) 100144doi:https://doi.org/10.1016/j.adapen.2023.100144

    work page doi:10.1016/j.adapen.2023.100144 2023

  16. [18]

    C. M. Calama-Gonz´alez, P. Symonds, G. Petrou, R. Su´arez, ´Angel Luis Le´on-Rodr´ıguez, Bayesian calibration of building energy models for un- certainty analysis through test cells monitoring, Applied Energy 282 (2021) 116118. doi:https://doi.org/10.1016/j.apenergy.2020. 116118

    work page doi:10.1016/j.apenergy.2020 2021

  17. [19]

    Martinez-Viol, E

    V . Martinez-Viol, E. M. Urbano, M. Delgado-Prieto, L. Romeral, Au- tomatic model calibration for coupled hvac and building dynamics us- ing modelica and bayesian optimization, Building and Environment 226 (2022) 109693. doi:https://doi.org/10.1016/j.buildenv.2022. 109693

    work page doi:10.1016/j.buildenv.2022 2022

  18. [20]

    S. Zhan, G. Wichern, C. Laughman, A. Chong, A. Chakrabarty, Calibrating building simulation models using multi-source datasets and meta-learned bayesian optimization, Energy and Buildings 270 (2022) 112278. doi: https://doi.org/10.1016/j.enbuild.2022.112278

    work page doi:10.1016/j.enbuild.2022.112278 2022

  19. [21]

    Chakrabarty, E

    A. Chakrabarty, E. Maddalena, H. Qiao, C. Laughman, Scalable bayesian optimization for model calibration: Case study on coupled building and hvac dynamics, Energy and Buildings 253 (2021) 111460. doi:https: //doi.org/10.1016/j.enbuild.2021.111460

    work page doi:10.1016/j.enbuild.2021.111460 2021

  20. [22]

    Oldewurtel, A

    F. Oldewurtel, A. Parisio, C. Jones, M. Morari, D. Gyalistras, M. Gwerder, V . Stauch, B. Lehmann, K. Wirth, Energy efficient building climate control using stochastic model predictive control and weather predictions, Pro- ceedings of the 2010 American Control Conference, ACC 2010 (2010) 5100 – 5105doi:10.1109/ACC.2010.5530680

    work page doi:10.1109/acc.2010.5530680 2010

  21. [23]

    J. Joe, P. Karava, A model predictive control strategy to optimize the per- formance of radiant floor heating and cooling systems in office buildings, Applied Energy 245 (2019) 65–77. doi:https://doi.org/10.1016/ j.apenergy.2019.03.209

    work page 2019

  22. [24]

    ENGIE, Engie website guidance, https://www.engie.be/nl/ elektriciteit-gas/, accessed: 2024-11-27

    work page 2024

  23. [25]

    selectandswitch, Select and switch website, https:// selectandswitch.com.au/, accessed: 2024-11-27

    work page 2024

  24. [26]

    Y . Cui, Z. Zhu, X. Zhao, Z. Li, Energy schedule setting based on clus- tering algorithm and pattern recognition for non-residential buildings electricity energy consumption, Sustainability 15 (11) (2023). doi: 10.3390/su15118750. URL https://www.mdpi.com/2071-1050/15/11/8750

    work page doi:10.3390/su15118750 2023

  25. [28]

    X. Lin, Q. Guo, D. Yuan, M. Gao, Bayesian optimization framework for hvac system control, Buildings 13 (2) (2023). doi:10.3390/ buildings13020314. URL https://www.mdpi.com/2075-5309/13/2/314

    work page 2023

  26. [29]

    Q. Lu, R. Kumar, V . M. Zavala, Mpc controller tuning using bayesian optimization techniques (2021). arXiv:2009.14175

    work page arXiv 2021

  27. [30]

    Q. Lu, L. D. Gonz ´alez, R. Kumar, V . M. Zavala, Bayesian optimiza- tion with reference models: A case study in mpc for hvac central plants, Computers & Chemical Engineering 154 (2021) 107491. doi:https: //doi.org/10.1016/j.compchemeng.2021.107491

    work page doi:10.1016/j.compchemeng.2021.107491 2021

  28. [31]

    H. Ren, D. ce Gao, Z. Ma, S. Zhang, Y . Sun, Data-driven surrogate optimization for deploying heterogeneous multi-energy storage to improve demand response performance at building cluster level, Applied Energy 356 (2024) 122312. doi:https://doi.org/10.1016/j.apenergy. 2023.122312

    work page doi:10.1016/j.apenergy 2024

  29. [32]

    Z. Deng, M. Chen, H. Wang, Q. Chen, Performance-oriented model learn- ing and model predictive control for pemfc air supply system, Interna- tional Journal of Hydrogen Energy 64 (2024) 339–348. doi:https: //doi.org/10.1016/j.ijhydene.2024.01.351

    work page doi:10.1016/j.ijhydene.2024.01.351 2024

  30. [33]

    W. Xu, Y . Jiang, B. Svetozarevic, C. Jones, Constrained efficient global op- timization of expensive black-box functions, in: International Conference on Machine Learning, PMLR, 2023, pp. 38485–38498

    work page 2023

  31. [34]

    M. W. Grieves, J. H. Vickers, Digital twin: Mitigating unpredictable, undesirable emergent behavior in complex systems, 2017. URL https://api.semanticscholar.org/CorpusID:14945167

    work page 2017

  32. [35]

    D. Blum, J. Arroyo, S. Huang, J. Drgo ˇna, F. Jorissen, H. T. Wal- num, Y . Chen, K. Benne, D. Vrabie, M. Wetter, L. Helsen, Building optimization testing framework (boptest) for simulation-based bench- marking of control strategies in buildings, Journal of Building Per- formance Simulation 14 (5) (2021) 586–610. arXiv:https://doi. org/10.1080/19401493.20...

    work page doi:10.1080/19401493.2021.1986574 2021

  33. [36]

    B¨unning, B

    F. B¨unning, B. Huber, A. Schalbetter, A. Aboudonia, M. Hudoba de Badyn, P. Heer, R. S. Smith, J. Lygeros, Physics-informed linear regression is competitive with two machine learning methods in residential building MPC, Appl. Energy 310 (118491) (2022) 118491

    work page 2022

  34. [37]

    Masset, R

    J. Drgoˇna, J. Arroyo, I. Cupeiro Figueroa, D. Blum, K. Arendt, D. Kim, E. P. Oll´e, J. Oravec, M. Wetter, D. L. Vrabie, L. Helsen, All you need to know about model predictive control for buildings, Annual Reviews in Control 50 (2020) 190–232. doi:https://doi.org/10.1016/j. arcontrol.2020.09.001

    work page doi:10.1016/j 2020

  35. [38]

    Bolusani, M

    S. Bolusani, M. Besan c ¸on, K. Bestuzheva, A. Chmiela, J. Dion ´ısio, T. Donkiewicz, J. van Doornmalen, L. Eifler, M. Ghannam, A. Gleixner, C. Graczyk, K. Halbig, I. Hedtke, A. Hoen, C. Hojny, R. van der Hulst, D. Kamp, T. Koch, K. Kofler, J. Lentz, J. Manns, G. Mexi, E. M ¨uhmer, M. E. Pfetsch, F. Schl¨osser, F. Serrano, Y . Shinano, M. Turner, S. Viger...

    work page 2024

  36. [39]

    Diamond, S

    S. Diamond, S. Boyd, CVXPY: A Python-embedded modeling language for convex optimization, Journal of Machine Learning Research 17 (83) (2016) 1–5

    work page 2016

  37. [40]

    MOSEK ApS, MOSEK Optimization Tools,https://www.mosek.com/, version 10.1.17 (2023)

    work page 2023

  38. [41]

    Engie website electricity contracts, https://www.engie.be/nl/ energie/elektriciteit-gas/prijzen-voorwaarden/ , accessed: 2024-03-22 (2024)

    work page 2024

  39. [42]

    P. O. Fanger, Thermal comfort: Analysis and applications in environmental engineering, Danish Technical Press, 1970

    work page 1970

  40. [43]

    En 15251:2007, indoor environmental input parameters for design and assessment of energy performance of buildings addressing indoor air quality, thermal environment, lighting and acoustics

    work page 2007

  41. [44]

    Tartarini, S

    F. Tartarini, S. Schiavon, pythermalcomfort: A python package for thermal comfort research, SoftwareX 12 (100578) (2020) 100578. 15

    work page 2020