pith. sign in

arxiv: 1906.12086 · v1 · pith:I2KU3FLTnew · submitted 2019-06-28 · 💻 cs.LG · cs.SY· eess.SY· stat.ML

Safe Contextual Bayesian Optimization for Sustainable Room Temperature PID Control Tuning

Marcello Fiducioso , Sebastian Curi , Benedikt Schumacher , Markus Gwerder , Andreas Krause This is my paper

Pith reviewed 2026-05-25 13:45 UTC · model grok-4.3

classification 💻 cs.LG cs.SYeess.SYstat.ML
keywords Bayesian optimizationPID tuningHVACsafe optimizationenergy efficiencycontextual optimizationtemperature control
0
0 comments X

The pith

Safe contextual Bayesian optimization tunes PID parameters for room temperature control, achieving 32% lower costs while enforcing comfort constraints.

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

The paper shows how safe contextual Bayesian optimization can automatically adjust the parameters of a PID controller that regulates room temperature in an HVAC system. This tuning process runs without human input and incorporates safety rules to prevent discomfort for occupants during the search for better settings. Buildings use 20 to 40 percent of national energy, with nearly half of that going to HVAC, so even modest efficiency gains matter for cost and emissions. The reported outcome is a 32 percent cost reduction relative to the existing controller while still meeting safety and comfort requirements.

Core claim

Safe contextual Bayesian optimization applied to PID tuning for a single room temperature loop delivers a 32 percent cost reduction against the current fixed setting, all while the safety constraints keep indoor conditions within acceptable comfort bounds for occupants.

What carries the argument

Safe contextual Bayesian optimization, which augments standard Bayesian optimization with both contextual variables and explicit safety constraints to select safe PID parameter values during online operation.

If this is right

  • Direct improvements to the specific room control loop and lower commissioning costs for that loop.
  • The same approach can be applied at other HVAC levels for additional energy and operational savings.
  • It supplies a concrete case of optimization techniques supporting environmental goals in building systems.

Where Pith is reading between the lines

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

  • Comparable safe-tuning methods could transfer to other building subsystems or industrial processes that require parameter adjustment under hard safety limits.
  • Repeated use over time might reveal whether further incremental gains beyond the initial 32 percent are possible once the controller is in steady operation.

Load-bearing premise

The safety model and constraint formulation inside the contextual Bayesian optimizer are sufficient to guarantee real-world comfort bounds during online tuning.

What would settle it

A physical deployment in which the learned PID settings produce measured room temperatures outside the declared comfort range or fail to deliver the claimed cost reduction.

Figures

Figures reproduced from arXiv: 1906.12086 by Andreas Krause, Benedikt Schumacher, Marcello Fiducioso, Markus Gwerder, Sebastian Curi.

Figure 1
Figure 1. Figure 1: Simplified plant schematic. The controller measures the [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Cumulative average cost of the different algorithms. For [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Parameters safe set in light blue and unsafe set in white. [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 2
Figure 2. Figure 2: Nevertheless, the global optimum is inside the safe [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 4
Figure 4. Figure 4: Optimal controller parameters as a function of the outside [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
read the original abstract

We tune one of the most common heating, ventilation, and air conditioning (HVAC) control loops, namely the temperature control of a room. For economical and environmental reasons, it is of prime importance to optimize the performance of this system. Buildings account from 20 to 40% of a country energy consumption, and almost 50% of it comes from HVAC systems. Scenario projections predict a 30% decrease in heating consumption by 2050 due to efficiency increase. Advanced control techniques can improve performance; however, the proportional-integral-derivative (PID) control is typically used due to its simplicity and overall performance. We use Safe Contextual Bayesian Optimization to optimize the PID parameters without human intervention. We reduce costs by 32% compared to the current PID controller setting while assuring safety and comfort to people in the room. The results of this work have an immediate impact on the room control loop performances and its related commissioning costs. Furthermore, this successful attempt paves the way for further use at different levels of HVAC systems, with promising energy, operational, and commissioning costs savings, and it is a practical demonstration of the positive effects that Artificial Intelligence can have on environmental sustainability.

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

1 major / 0 minor

Summary. The paper claims that Safe Contextual Bayesian Optimization can be applied to tune the parameters of a PID controller for room temperature regulation in an HVAC system. It reports a 32% reduction in costs relative to the existing controller settings, while maintaining safety and occupant comfort, with potential immediate impact on commissioning costs and broader applicability to other HVAC loops for energy and operational savings.

Significance. If the safety constraints are shown to hold under real-world disturbances and the 32% figure is reproducible with proper baselines and statistics, the work would provide a concrete demonstration of safe online optimization for sustainability. Buildings account for 20-40% of energy consumption with HVAC as a major contributor, so validated methods that reduce consumption without compromising comfort could have practical environmental impact. The absence of error bars, baseline comparisons, and safety verification details in the presented material prevents a full assessment of significance.

major comments (1)
  1. The central empirical claim (32% cost reduction while assuring safety and comfort) is stated in the abstract without error bars, baseline controller details, trial duration, or quantitative evidence that the safety model and constraint formulation prevented real-world comfort violations. This modeling assumption is load-bearing for the claim that costs can be reduced while guaranteeing comfort, yet no validation against occupancy effects or disturbances is described.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the constructive feedback on our application of Safe Contextual Bayesian Optimization to real-world PID tuning in HVAC systems. We agree that the central empirical claim requires stronger statistical support and explicit safety validation details to substantiate the reported cost savings while maintaining comfort. We address the major comment below and commit to revisions that improve the rigor of the presentation without altering the core experimental findings.

read point-by-point responses

  1. Referee: The central empirical claim (32% cost reduction while assuring safety and comfort) is stated in the abstract without error bars, baseline controller details, trial duration, or quantitative evidence that the safety model and constraint formulation prevented real-world comfort violations. This modeling assumption is load-bearing for the claim that costs can be reduced while guaranteeing comfort, yet no validation against occupancy effects or disturbances is described.

    Authors: We agree with the referee that the abstract presents the 32% figure without accompanying statistical details, which limits assessment of reproducibility and robustness. The full manuscript (Section 4) specifies the baseline as the pre-existing PID settings in the building management system and describes a multi-week real-world deployment in an occupied room. However, error bars are indeed not reported for the cost metric, and no table or plot quantifies comfort violations (e.g., time outside temperature bounds) or explicitly validates that the safety constraints averted issues under occupancy or external disturbances. We will revise the abstract, add error bars derived from the optimization trajectory, expand the baseline description, and include a new figure or subsection with quantitative safety metrics from the experiment. The contextual formulation incorporates variables that partially account for environmental variation, but dedicated ablation on occupancy disturbances is absent; we will add an explicit limitations paragraph acknowledging this. These changes will be incorporated in the revised manuscript. revision: yes

Circularity Check

0 steps flagged

No significant circularity; result is direct empirical measurement

full rationale

The paper applies an existing Safe Contextual Bayesian Optimization method to tune PID parameters on a physical HVAC system and reports a measured 32% cost reduction from hardware trials. No derivation chain, first-principles predictions, or fitted quantities are presented that could reduce to their own inputs by construction. The central claim is an observed outcome on real equipment rather than a self-referential or statistically forced prediction, so the result is self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review; no free parameters, axioms, or invented entities are visible in the provided text.

pith-pipeline@v0.9.0 · 5756 in / 971 out tokens · 25929 ms · 2026-05-25T13:45:31.759308+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

27 extracted references · 27 canonical work pages · 1 internal anchor

  1. [1]

    Bayesian optimization for maximum power point tracking in pho- tovoltaic power plants

    [Abdelrahman et al., 2016] Hany Abdelrahman, Felix Berkenkamp, Jan Poland, and Andreas Krause. Bayesian optimization for maximum power point tracking in pho- tovoltaic power plants. In European Control Conference (ECC), pages 2078–2083,

    work page 2016

  2. [2]

    Theory and applications of hvac control systems–a review of model predictive control

    [Afram and Janabi-Sharifi, 2014] Abdul Afram and Farrokh Janabi-Sharifi. Theory and applications of hvac control systems–a review of model predictive control. Building and Environment, 72:343–355,

    work page 2014

  3. [3]

    Funda- mentals of heat and mass transfer

    [Bergman et al., 2011] Theodore L Bergman, Frank P Incr- opera, David P DeWitt, and Adrienne S Lavine. Funda- mentals of heat and mass transfer . John Wiley & Sons,

    work page 2011

  4. [4]

    Bayesian optimization with safety constraints: safe and automatic parameter tuning in robotics

    [Berkenkamp et al., 2016] Felix Berkenkamp, Andreas Krause, and Angela P Schoellig. Bayesian optimization with safety constraints: safe and automatic parameter tuning in robotics. arXiv:1602.04450,

    work page arXiv 2016

  5. [5]

    Advanced controller auto-tuning and its application in hvac systems

    [Bi et al., 2000] Qiang Bi, Wen-Jian Cai, Qing-Guo Wang, Chang-Chieh Hang, Eng-Lock Lee, Yong Sun, Ke-Dian Liu, Yong Zhang, and Biao Zou. Advanced controller auto-tuning and its application in hvac systems. Control Engineering Practice, 8(6):633–644,

    work page 2000

  6. [6]

    Convergence rates of effi- cient global optimization algorithms

    [Bull, 2011] Adam D Bull. Convergence rates of effi- cient global optimization algorithms. Journal of Machine Learning Research, 12(Oct):2879–2904,

    work page 2011

  7. [7]

    An experimental comparison of bayesian optimization for bipedal locomo- tion

    [Calandra et al., 2014] Roberto Calandra, Andr ´e Seyfarth, Jan Peters, and Marc Peter Deisenroth. An experimental comparison of bayesian optimization for bipedal locomo- tion. In ICRA, pages 1951–1958,

    work page 2014

  8. [8]

    Control of sys- tems with costs related to switching: applications to air-condition systems

    [Deng et al., 2009] Honglian Deng, Lars Larsen, Jakob Stoustrup, and Henrik Rasmussen. Control of sys- tems with costs related to switching: applications to air-condition systems. In 2009 IEEE Control Applica- tions,(CCA) & Intelligent Control,(ISIC), pages 554–559. IEEE,

    work page 2009

  9. [9]

    Increasing customer value of industrial control performance monitoring-honeywell’s experience

    [Desborough and Miller, 2002] Lane Desborough and Randy Miller. Increasing customer value of industrial control performance monitoring-honeywell’s experience. In AIChE symposium series, number 326, pages 169–189. American Institute of Chemical Engineers,

    work page 2002

  10. [10]

    Deep- mind ai reduces google data centre cooling bill by 40%

    [Evans and Gao, 2016] Richard Evans and Jim Gao. Deep- mind ai reduces google data centre cooling bill by 40%. DeepMind blog, 20,

    work page 2016

  11. [11]

    Multizone building with vav air- conditioning system simulation for evaluation and test of control systems

    [Gao et al., 2007] Tianyun Gao, Benedikt Schumacher, Wal- ter Hegetschweiler, Markus Gwerder, Michael Tschanz, and Marcel Walti. Multizone building with vav air- conditioning system simulation for evaluation and test of control systems. 09

    work page 2007

  12. [12]

    Model predictive control: theory and practice—a survey

    [Garcia et al., 1989] Carlos E Garcia, David M Prett, and Manfred Morari. Model predictive control: theory and practice—a survey. Automatica, 25(3):335–348,

    work page 1989

  13. [13]

    Bayesian Optimization with Unknown Constraints

    [Gelbart et al., 2014] Michael A Gelbart, Jasper Snoek, and Ryan P Adams. Bayesian optimization with unknown con- straints. arXiv:1403.5607,

    work page internal anchor Pith review Pith/arXiv arXiv 2014

  14. [14]

    Contextual gaussian process bandit optimization

    [Krause and Ong, 2011] Andreas Krause and Cheng S Ong. Contextual gaussian process bandit optimization. In Ad- vances in Neural Information Processing Systems , pages 2447–2455,

    work page 2011

  15. [15]

    Linear optimal control systems, volume

    [Kwakernaak and Sivan, 1972] Huibert Kwakernaak and Raphael Sivan. Linear optimal control systems, volume

    work page 1972

  16. [16]

    Support vector regression and ant colony op- timization for hvac cooling load prediction

    [Lixing et al., 2010] Ding Lixing, Lv Jinhu, Li Xuemei, and Li Lanlan. Support vector regression and ant colony op- timization for hvac cooling load prediction. In Computer Communication Control and Automation (3CA), volume 1, pages 537–541. IEEE,

    work page 2010

  17. [17]

    Virtual vs

    [Marco et al., 2017] Alonso Marco, Felix Berkenkamp, Philipp Hennig, Angela P Schoellig, Andreas Krause, Ste- fan Schaal, and Sebastian Trimpe. Virtual vs. real: Trad- ing off simulations and physical experiments in reinforce- ment learning with bayesian optimization. In ICRA, pages 1557–1563. IEEE,

    work page 2017

  18. [18]

    Handbook of PI and PID controller tuning rules

    [O’Dwyer, 2006] Aidan O’Dwyer. Handbook of PI and PID controller tuning rules. World Scientific,

    work page 2006

  19. [19]

    Energy efficient building climate control using stochastic model predictive control and weather predictions

    [Oldewurtel et al., 2010] Frauke Oldewurtel, Alessandra Parisio, Colin Jones, Manfred Morari, Dimitrios Gyalis- tras, Markus Gwerder, Vanessa Stauch, Beat Lehmann, and Katharina Wirth. Energy efficient building climate control using stochastic model predictive control and weather predictions. In Proceedings of the 2010 American control conference, pages 5100–5105,

    work page 2010

  20. [20]

    A review on buildings energy con- sumption information

    [P´erez-Lombard et al., 2008] Luis P´erez-Lombard, Jos´e Or- tiz, and Christine Pout. A review on buildings energy con- sumption information. Energy and buildings, 40(3):394– 398,

    work page 2008

  21. [21]

    Gaussian pro- cesses in machine learning

    [Rasmussen, 2004] Carl Edward Rasmussen. Gaussian pro- cesses in machine learning. In Advanced lectures on ma- chine learning, pages 63–71. Springer,

    work page 2004

  22. [22]

    A new pattern recognition adap- tive controller with application to hvac systems

    [Seem, 1998] John E Seem. A new pattern recognition adap- tive controller with application to hvac systems. Automat- ica, 34(8):969–982,

    work page 1998

  23. [23]

    Practical bayesian optimization of ma- chine learning algorithms

    [Snoek et al., 2012] Jasper Snoek, Hugo Larochelle, and Ryan P Adams. Practical bayesian optimization of ma- chine learning algorithms. In Advances in neural informa- tion processing systems, pages 2951–2959,

    work page 2012

  24. [24]

    Gaussian process bandits without regret: An experimental design approach

    [Srinivas et al., 2009] Niranjan Srinivas, Sham M Kakade, Matthias Seeger, and Andreas Krause. Gaussian process bandits without regret: An experimental design approach. Technical report,

    work page 2009

  25. [25]

    Safe exploration for optimization with gaussian processes

    [Sui et al., 2015] Yanan Sui, Alkis Gotovos, Joel Burdick, and Andreas Krause. Safe exploration for optimization with gaussian processes. In International Conference on Machine Learning, pages 997–1005,

    work page 2015

  26. [26]

    Robust and optimal control , vol- ume

    [Zhou et al., 1996] Kemin Zhou, John Comstock Doyle, Keith Glover, et al. Robust and optimal control , vol- ume

    work page 1996

  27. [27]

    Optimum settings for automatic controllers

    [Ziegler and Nichols, 1942] John G Ziegler and Nathaniel B Nichols. Optimum settings for automatic controllers. trans. ASME, 64(11), 1942

    work page 1942