Pith. sign in

REVIEW 3 major objections 6 minor 43 references

A Data-driven Predictive Control Architecture for Train Thermal Energy Management

T0 review · 3 major / 6 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 Adding a data-driven predictive setpoint controller to train HVAC cuts measured energy use by 10-35%.

desk verdict A credible first deployment of DDPC on real trains, but the headline 10-35% savings is not causally identified by four single-pair, non-moving experiments. read the letter →

arxiv 2506.09187 v1 pith:T646SBS5 submitted 2025-06-10 eess.SY cs.SY

Ahmed Aboudonia , Johannes Estermann , Keith Moffat , Manfred Morari , John Lygeros This is my paper
classification eess.SYcs.SY
keywords trainHVACdata-drivenpredictivecontrolTransientPredictorenergyefficiencytemperaturesetpointoptimizationmodelRegio-DostotrainsSwissFederalRailways
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 claims that a comparatively small addition to an existing train climate-control system—a middle-layer predictive controller that chooses the HVAC temperature setpoint—can cut HVAC energy use by roughly a tenth to a third. The authors build a linear multistep prediction model from a few months of logged data on a handful of Swiss regional train coaches and embed it in a model predictive control loop that nudges the setpoint within the ±2°C band a driver is already allowed to use. In four field experiments on empty, stationary trains, the activated controller used 11.65% to 34.94% less energy than the rule-based baseline while keeping temperature-bound violations small. Because HVAC is the second-largest energy consumer in these trains after traction, a successful retrofit layer would deliver fleet-wide savings without new hardware.

What carries the argument

The central object is the Transient Predictor, a data-driven estimator that fits a causal multistep prediction model from historical input-output-disturbance trajectories using an LQ decomposition of concatenated Hankel matrices. The resulting predictor supplies the controller with a linear forecast of the upper, middle, and lower deck temperatures over the horizon, so future outputs at each step depend only on data available up to that step. The controller's cost then does two things at once: an energy term pulls the reference temperature toward the open-loop optimal boundary $T_{\mathrm{opt}}(k|t)$ computed by simulating the coach with the HVAC off, and a temperature term that, by the paper's Theorem 1, is equivalent to penalizing both the deviation of the average deck temperature from $T_{\mathrm{opt}}$ and the spatial spread among decks. The $\pm 2^\circ\mathrm{C}$ setpoint constraint and the slew-rate limit tie the layer to the authority a driver already has, so the existing rule-based and third-party tracking controllers remain in place.

What would settle it

Run the same A/B comparison on occupied trains in normal scheduled service with coaches randomly assigned to activated or baseline control, logging power and occupancy per trip: if the average energy difference after accounting for occupancy and weather is statistically indistinguishable from zero, or if the savings appear only when temperature constraints are violated, then the 10-35% range does not transfer to revenue service.

Watch

Extended reading notes

Core claim

The central claim is that the existing two-level train climate architecture can be made substantially more efficient by inserting a data-driven predictive layer between the rule-based setpoint generator and the low-level tracking controller. The layer forecasts the three deck temperatures over a 30-minute horizon with a multistep predictor built from real coach data, then solves a quadratic program every 5 minutes to choose a reference temperature that keeps each deck within a prespecified band around the rule-based setpoint, respects the driver's ±2°C override range and setpoint slew limits, and pushes the setpoint toward the energy-optimal boundary—the lower bound during heating and the upper bound during cooling. The authors report that in all four experiments the activated coach consumed less power than its paired baseline coach, with steady-state savings of 28.03%, 20.11%, 34.94%, and 11.65%; the three cooling experiments show a clear trade-off between energy savings and temperature-constraint violations, while the heating experiment achieves large savings with small violations.

Load-bearing premise

The load-bearing premise is that energy savings measured on empty, stationary trains during scheduled breaks, comparing one activated coach against one paired baseline coach, will carry over to occupied, moving trains in normal service where coach-to-coach variability, passenger heat, door openings, and solar exposure vary.

Editorial extensions

If this is right

  • Fleet-wide deployment on Regio-Dosto trains could reduce HVAC energy use by roughly 10-35% without replacing the rule-based controller or the third-party tracking controller, since the added layer only modifies the setpoint within the driver's existing authority.
  • Operators can trade energy savings against comfort by tuning cost weights and the conservative temperature bound: the three cooling experiments show higher savings accompanied by larger average hourly violations.
  • Because the prediction model is trained on data from a few coaches and evaluated on separate coaches, the architecture offers a path to rollout without per-coach system identification.
  • The controller's setpoint automatically follows the weather, moving to the lower bound for heating in cold conditions and to the upper bound for cooling in warm conditions, which is what produces the savings in both seasons.

Reading between the lines

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

  • Editorial inference: the same setpoint-nudging layer could transfer to other vehicles or buildings whose HVAC controller is a sealed unit, because it only needs a setpoint interface, temperature measurements, and weather forecasts.
  • Editorial inference: the field savings are likely upper-bound estimates for revenue service; empty, stationary coaches lack passenger heat loads, door openings, and the solar and speed variations of moving trains, so occupied operation would probably compress the savings range unless the controller is retuned with occupancy forecasts.
  • Editorial inference: a direct testable extension is to use the same predictor with occupancy as a measured disturbance and compare passenger comfort indicators, not just temperature bounds, against the rule-based baseline on occupied trains.
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.

Referee Report

3 major / 6 minor

Summary. The paper proposes adding a data-driven predictive control (DDPC) layer to the existing rule-based HVAC control architecture of Swiss Regio-Dosto trains. The layer uses a multistep linear prediction model built with the Transient Predictor, and an MPC formulation that chooses the HVAC setpoint to minimize energy while respecting comfort bounds. The authors validate the prediction model on held-out coaches and report simulation results, then describe four field experiments comparing one coach with DDPC activated against one coach with DDPC deactivated, claiming energy savings between 10% and 35%.

Significance. If the claimed savings are real, the work has clear practical value: HVAC is the second-largest energy consumer in these trains, the proposed layer is complementary to the existing architecture, and the study includes genuine deployments on real trains. The paper also has concrete strengths: the prediction model is validated on held-out coaches with reported MAE values up to 0.25 °C over 60 minutes, energy consumption is measured with external power loggers rather than derived from the fitted model, and the method is evaluated in both simulation and field settings. However, the headline 10–35% savings claim rests on a small number of single-pair comparisons without crossover or replication, and the experiments are conducted on empty, non-moving trains during scheduled breaks. The significance of the contribution therefore depends on either substantially stronger experimental evidence or a carefully qualified claim restricted to the specific conditions tested.

major comments (3)
  1. [Section V, Table II] The central energy-savings claim is not causally identified. Each of the four experiments compares one coach with DDPC activated against one coach with DDPC deactivated, on the same train, without crossover, repetition, or randomization. The paper itself acknowledges coach-to-coach variability in Section VI, but Table II reports only point estimates (28.03%, 20.11%, 34.94%, 11.65%) with no confidence intervals or variability measures. Because the experiments are on empty, non-moving trains, occupancy, door effects, solar exposure, and train-motion disturbances—which the paper identifies as important—are absent. To support the abstract's 10–35% claim, the authors should either add a within-coach crossover baseline or replicate the comparison over multiple coaches per condition and report the spread; otherwise the claim should be explicitly downgraded to a preliminary case-study observation.
  2. [Section V, Experiment 1 and Table II] The handling of connection losses in the energy bookkeeping is underspecified. The text states that whenever connection to either HVAC system is lost, the power consumption is set to zero, and that the savings calculation starts at 14:00. In Experiment 1, the deactivated DDPC coach loses telemetry at 13:45 and the activated DDPC coach loses connection around 17:50. Table II says affected intervals are excluded, but it is not stated whether zeroed samples that fall inside the averaging window are excluded or included. This choice can materially affect the 28.03% figure, since zeroing the deactivated coach's power would inflate savings while zeroing the activated coach's power would deflate them. The authors should state the exact sample-inclusion rule and report sensitivity to the two conventions.
  3. [Section III-C] The 'optimal bound' T_opt is computed by simulating the open-loop thermal dynamics with Q_hvac = 0 using the authors' SBB simulator, and then projecting the resulting temperature trajectory onto the admissible interval. This heuristic is presented without validation that it actually identifies the energy-minimizing boundary over the prediction horizon. Since the cost terms J_u and J_y drive the controller toward T_opt, errors in this heuristic directly affect the setpoint choices and hence the measured savings. A simulation study comparing T_opt against the actual optimal boundary, or against the closed-loop energy consumption of always choosing the upper versus the lower bound, would strengthen the architecture independently of the field experiments.
minor comments (6)
  1. [Abstract and Conclusion] The abstract states savings between 10% and 35%, while the conclusion states between 10% and 30% and Table II includes 34.94%. These should be made consistent.
  2. [Section II-B] There are minor typos: 'respecetively' and 'equiped' should be corrected.
  3. [Figure 5] The caption refers to a 'thick red line' marking the parameter set used later, but the described red line is not clearly visible or labeled in the figure as reproduced; please clarify the figure or the caption.
  4. [Section III-A] The symbol N is used both for the number of trajectories and for the total number of columns in the concatenated Hankel matrix; this overloading is confusing and should be resolved with distinct notation.
  5. [Section V] The bar plot of 'percentage of energy savings every 30 minutes' is described qualitatively, but the text does not define how the per-interval savings percentage is computed; a short formal definition would improve reproducibility.
  6. [Table II] The table reports averages without any measure of dispersion or the duration of each evaluation window; at minimum the length of the steady-state window per experiment should be stated.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: savings are field-measured and the predictor is validated on held-out coaches.

full rationale

The paper's central claim, 10-35% energy savings, rests on external power-logger measurements on real coaches under the proposed DDPC architecture versus a rule-based baseline, not on the fitted prediction model itself. The multistep predictor is trained on some coaches and explicitly evaluated on held-out coaches in Section IV-A, reporting a mean absolute error of at most 0.25 C over 60 minutes on a validation dataset; this is an out-of-sample check rather than a reuse of training data. The T_opt heuristic used in the cost function is a control-design choice computed from the SBB simulator, but it is not the quantity in which the claimed savings are defined or measured. The paper does cite the authors' own Transient Predictor [38] for the model structure, and Algorithm 1 restates the computation; however, the model is independently validated on held-out data and the final energy claims are empirical, so this self-citation is not load-bearing in a circular sense. Section VI honestly flags that occupied-train comparisons require a more rigorous methodology due to coach-to-coach variability; that is an external-validity limitation, not a definitional equivalence. The lack of crossover in the four field experiments is a causal-identification concern, but it does not make the derivation circular: the measured savings are not constructed from the model's fitted parameters or from the paper's own equations. The derivation chain is therefore self-contained.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The central savings claim rests on a fitted data-driven predictor, a simulator-derived energy target, assumed forecast accuracy, and extrapolation from unoccupied experiments. No new physical entities are introduced; the controller introduces tuning parameters and a heuristic T_opt that carry much of the burden.

free parameters (4)
  • Cost weights sigma, tau, gamma (Eq. 9) = not reported; varied per experiment
    Hand-tuned weights determine the balance between energy (J_u), comfort (J_y), slack (J_eps), and setpoint rate (J_Delta_u). Experiment 3 increases the energy weight, Experiment 2 uses a more conservative bound; the savings figures depend on these choices.
  • Prediction horizon, lead-in, sampling time = T=30 min, rho=12, sampling=5 min
    Chosen by inspection of Figure 5 for the implemented controller; prediction accuracy varies with these parameters and 5-minute sampling is a communication and computation compromise.
  • Comfort bound T_max and setpoint rate limit Delta_T_max = SBB bound +/-1 deg C; Delta_T_max not reported
    Equations (4) and (7) define comfort and indirectly limit HVAC heat flow; Experiment 2 adds a conservative upper bound. Values materially affect constraint violations and savings.
  • Data-driven predictor matrices Phi and H = unknown coefficients
    Estimated from real coach trajectories via LQ decomposition; setpoint decisions in the DDPC layer are a function of these fitted matrices.
assumptions (5)
  • domain assumption A linear multistep predictor is sufficient to represent train coach thermal dynamics over a 30-minute horizon.
    Section III-B restricts the model to linear even though the simulator includes nonlinearities such as ground radiation; Figure 5 shows MAE up to 0.25 deg C, so the assumption is partially tested but not proven for all regimes.
  • ad hoc to paper The simulator-in-the-loop target T_opt identifies the energy-optimal temperature bound.
    Section III-C computes T_opt by simulating open-loop dynamics with Q_hvac=0 and projecting onto the comfort interval, then uses it as the energy surrogate; no equivalence proof links this target to actual HVAC energy minimization.
  • domain assumption Future weather, solar, and occupancy disturbance forecasts are available with sufficient accuracy.
    The predictor consumes future disturbance trajectories d_f; experiments use real-time Meteomatics weather. The paper lists forecast uncertainty and robust or stochastic extensions as future work.
  • domain assumption Energy savings measured on empty, parked coaches transfer to occupied, moving service.
    Section V states safety-limited experiments on empty non-moving trains during scheduled breaks; Section VI says occupied scenarios require a more rigorous methodology. The headline 10-35% range assumes this transferability.
  • domain assumption The SBB simulator used to generate T_opt is faithful to real train thermal behavior.
    T_opt depends on the simulator described in Section II and an unpublished SBB feasibility study [39]; no public validation of the simulator against the test coaches is provided.

how reviews work

0 comments
Cite this review

Pith. "Pith review of A Data-driven Predictive Control Architecture for Train Thermal Energy Management." pith.science (2026). https://pith.science/paper/T646SBS5

@misc{pith2026250609187,
  author       = {Pith},
  title        = {Pith review of: A Data-driven Predictive Control Architecture for Train Thermal Energy Management},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/T646SBS5}},
  note         = {Machine review of arXiv:2506.09187}
}
read the original abstract

We aim to improve the energy efficiency of train climate control architectures, with a focus on a specific class of regional trains operating throughout Switzerland, especially in Zurich and Geneva. Heating, Ventilation, and Air Conditioning (HVAC) systems represent the second largest energy consumer in these trains after traction. The current architecture comprises a high-level rule-based controller and a low-level tracking controller. To improve train energy efficiency, we propose adding a middle data-driven predictive control layer aimed at minimizing HVAC energy consumption while maintaining passenger comfort. The scheme incorporates a multistep prediction model developed using real-world data collected from a limited number of train coaches. To validate the effectiveness of the proposed architecture, we conduct multiple experiments on a separate set of train coaches; our results suggest energy savings between 10% and 35% with respect to the current architecture.

Figures

Figures reproduced from arXiv: 2506.09187 by the authors.

Figure 1
Figure 1. A schematic diagram of the considered coach. [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. A block diagram of the SBB simulator. the same setpoint Tref , each deck is controlled individually and hence, we define Troom = [T up room T mid room T low room] ⊤ and Q˙ hvac = [Q˙ up hvac Q˙ mid hvac Q˙ low hvac] ⊤ where the superscripts up, mid and low refer, respecetively, to the upper, middle and lower decks shown in [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Thermal Dynamics [39]. distribution block describes the relation between Q˙ in and Q˙ act that is given by, Q˙ act = ΛQ˙ in, where Λ is a matrix with each column summing up to 1 and whose i-th row and j-th column entry describes the fraction of the contribution of the j-th element in Q˙ in to the i-th element in Q˙ act. The coach sub-block models the thermal behavior of each part in the coach illustrated in [PITH_F… view at source ↗
Figures from the paper (8 more)
Figure 4
Figure 4. Figure 4: Proposed control architecture. the error between the reference temperature Tref and the room temperature Troom. This helps to ensure that the output of the low-level tracking controller included in the HVAC System and Control block in [PITH_FULL_IMAGE:figures/full_fig…
Figure 6
Figure 6. Figure 6: The train starts its operation slightly after 04:00 on [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 5
Figure 5. Figure 5: Data driven model evaluation (Note: The thick red line in each plot refers to the set of parameters that we will use in the proposed controller). [PITH_FULL_IMAGE:figures/full_fig_p009_5.png]
Figure 6
Figure 6. Figure 6: DDPC evaluation in simulation: The DDPC scheme moves towards the lower bound in the beginning and end of the day and towards the upper bound [PITH_FULL_IMAGE:figures/full_fig_p009_6.png]
Figure 7
Figure 7. Figure 7: Cooling Experiment 1: The “Activated DDPC” scheme aims to minimize the energy consumption while respecting the temperature bounds by moving [PITH_FULL_IMAGE:figures/full_fig_p010_7.png]
Figure 8
Figure 8. Figure 8: Cooling Experiment 2: The “Activated DDPC” scheme consumes less energy in comparison to the “Deactivated DDPC” scheme as it approaches the [PITH_FULL_IMAGE:figures/full_fig_p011_8.png]
Figure 9
Figure 9. Figure 9: Cooling Experiment 3: The “Activated DDPC” scheme minimizes the energy consumption by moving towards the upper bound and violating the [PITH_FULL_IMAGE:figures/full_fig_p012_9.png]
Figure 10
Figure 10. Figure 10: Heating Experiment 4: The “Activated DDPC” scheme consumes less energy in comparison to the “Deactivated DDPC” by moving towards the [PITH_FULL_IMAGE:figures/full_fig_p013_10.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

43 extracted references · 40 canonical work pages

  1. [1]

    Theory and applications of hvac control systems–a review of model predictive control (mpc),

    A. Afram and F. Janabi-Sharifi, “Theory and applications of hvac control systems–a review of model predictive control (mpc),”Building and Environment, vol. 72, pp. 343–355, 2014

    work page 2014

  2. [2]

    A review on energy sav- ing strategies in industrial sector,

    E. Abdelaziz, R. Saidur, and S. Mekhilef, “A review on energy sav- ing strategies in industrial sector,”Renewable and sustainable energy reviews, vol. 15, no. 1, pp. 150–168, 2011

    work page 2011

  3. [3]

    A survey on optimal trajectory planning for train operations,

    Y . Wang, B. Ning, F. Cao, B. De Schutter, and T. J. Van den Boom, “A survey on optimal trajectory planning for train operations,” inPro- ceedings of 2011 IEEE International Conference on Service Operations, Logistics and Informatics, pp. 589–594, IEEE, 2011

    work page 2011

  4. [4]

    Hierarchical model predictive control for coordinated electric railway traction system energy manage- ment,

    H. Novak, V . Le ˇsi´c, and M. Va ˇsak, “Hierarchical model predictive control for coordinated electric railway traction system energy manage- ment,”IEEE Transactions on Intelligent Transportation Systems, vol. 20, no. 7, pp. 2715–2727, 2018

    work page 2018

  5. [5]

    Single-train trajectory optimization,

    S. Lu, S. Hillmansen, T. K. Ho, and C. Roberts, “Single-train trajectory optimization,”IEEE Transactions on Intelligent Transportation Systems, vol. 14, no. 2, pp. 743–750, 2013

    work page 2013

  6. [6]

    Energy-efficient locomotive operation for chinese mainline railways by fuzzy predic- tive control,

    Y . Bai, T. K. Ho, B. Mao, Y . Ding, and S. Chen, “Energy-efficient locomotive operation for chinese mainline railways by fuzzy predic- tive control,”IEEE Transactions on Intelligent Transportation Systems, vol. 15, no. 3, pp. 938–948, 2014

    work page 2014

  7. [7]

    Y . Wang, B. Ning, T. Van den Boom, B. De Schutter,et al.,Optimal trajectory planning and train scheduling for urban rail transit systems. Springer, 2016

    work page 2016

  8. [8]

    A cooperative scheduling model for timetable optimization in subway systems,

    X. Yang, X. Li, Z. Gao, H. Wang, and T. Tang, “A cooperative scheduling model for timetable optimization in subway systems,”IEEE Transactions on Intelligent Transportation Systems, vol. 14, no. 1, pp. 438–447, 2012

    work page 2012

Show all 43 references
  1. [9]

    An energy-efficient scheduling approach to improve the utilization of regenerative energy for metro systems,

    X. Yang, A. Chen, X. Li, B. Ning, and T. Tang, “An energy-efficient scheduling approach to improve the utilization of regenerative energy for metro systems,”Transportation Research Part C: Emerging Tech- JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2021 14 nologies, vo...

    2021

  2. [10]

    Hierarchical control and full-range dynamic performance optimization of the supercapacitor energy storage system in urban railway,

    F. Zhu, Z. Yang, H. Xia, and F. Lin, “Hierarchical control and full-range dynamic performance optimization of the supercapacitor energy storage system in urban railway,”IEEE Transactions on Industrial Electronics, vol. 65, no. 8, pp. 6646–6656, 2017

    2017

  3. [11]

    Multiport converters and ess on 3-kv dc railway lines: Case study for braking energy savings,

    A. Clerici, E. Tironi, and F. Castelli-Dezza, “Multiport converters and ess on 3-kv dc railway lines: Case study for braking energy savings,” IEEE Transactions on Industry Applications, vol. 54, no. 3, pp. 2740– 2750, 2018

    2018

  4. [12]

    Energy efficiency of railway vehicles,

    N. Vetterli, U.-P. MENTI, F. SIDLER, E. THALER, and G. ZWEIFEL, “Energy efficiency of railway vehicles,” inProceedings of International Conference CISBAT 2015 Future Buildings and Districts Sustainability from Nano to Urban Scale, pp. 955–960, LESO-PB, EPFL, 2015

    2015

  5. [13]

    J. B. Rawlings, D. Q. Mayne, M. Diehl,et al.,Model predictive control: theory, computation, and design, vol. 2. Nob Hill Publishing Madison, WI, 2017

    2017

  6. [14]

    Model predictive control,

    B. Kouvaritakis and M. Cannon, “Model predictive control,”Switzer- land: Springer International Publishing, vol. 38, pp. 13–56, 2016

    2016

  7. [15]

    Model predictive control: Review of the three decades of development,

    J. H. Lee, “Model predictive control: Review of the three decades of development,”International Journal of Control, Automation and Systems, vol. 9, pp. 415–424, 2011

    2011

  8. [16]

    Handling model uncertainty in model predictive control for energy efficient buildings,

    M. Maasoumy, M. Razmara, M. Shahbakhti, and A. S. Vincentelli, “Handling model uncertainty in model predictive control for energy efficient buildings,”Energy and Buildings, vol. 77, pp. 377–392, 2014

    2014

  9. [17]

    Optimal control of building HV AC systems in the presence of imperfect predictions,

    M. Maasoumy and A. Sangiovanni-Vincentelli, “Optimal control of building HV AC systems in the presence of imperfect predictions,” in Dynamic Systems and Control Conference, vol. 45301, pp. 257–266, American Society of Mechanical Engineers, 2012

    2012

  10. [18]

    Energy saving and indoor tem- perature control for an office building using tube-based robust model predictive control,

    Y . Gao, S. Miyata, and Y . Akashi, “Energy saving and indoor tem- perature control for an office building using tube-based robust model predictive control,”Applied Energy, vol. 341, p. 121106, 2023

    2023

  11. [19]

    A scenario-based distributed stochastic mpc for building temperature regulation,

    Y . Long, S. Liu, L. Xie, and K. H. Johansson, “A scenario-based distributed stochastic mpc for building temperature regulation,” in2014 IEEE international conference on automation science and engineering (CASE), pp. 1091–1096, IEEE, 2014

    2014

  12. [20]

    On the sample size of randomized mpc for chance-constrained systems with application to building climate control,

    X. Zhang, S. Grammatico, G. Schildbach, P. Goulart, and J. Lygeros, “On the sample size of randomized mpc for chance-constrained systems with application to building climate control,” in2014 european control conference (ECC), pp. 478–483, IEEE, 2014

    2014

  13. [21]

    Stochastic model predictive control for building hvac systems: Complexity and conservatism,

    Y . Ma, J. Matu ˇsko, and F. Borrelli, “Stochastic model predictive control for building hvac systems: Complexity and conservatism,”IEEE Trans- actions on Control Systems Technology, vol. 23, no. 1, pp. 101–116, 2014

    2014

  14. [22]

    Stochastic model predictive control for building climate control,

    F. Oldewurtel, C. N. Jones, A. Parisio, and M. Morari, “Stochastic model predictive control for building climate control,”IEEE Transactions on Control Systems Technology, vol. 22, no. 3, pp. 1198–1205, 2013

    2013

  15. [23]

    Self-adapting building models for model predictive control,

    S. Herzog, D. Atabay, J. Jungwirth, and V . Mikulovic, “Self-adapting building models for model predictive control,” inBuilding Simulation 2013, vol. 13, pp. 2489–2493, IBPSA, 2013

    2013

  16. [24]

    Savings through the use of adaptive predictive control of thermo-active building systems (tabs): A case study,

    M. Schmelas, T. Feldmann, and E. Bollin, “Savings through the use of adaptive predictive control of thermo-active building systems (tabs): A case study,”Applied Energy, vol. 199, pp. 294–309, 2017

    2017

  17. [25]

    Robust adaptive model predictive building climate control. ifac-papersonline, 50 (1): 1871–1876, 2017,

    M. Tanaskovic, D. Sturzenegger, R. Smith, and M. Morari, “Robust adaptive model predictive building climate control. ifac-papersonline, 50 (1): 1871–1876, 2017,” in20th IFAC World Congress

    2017

  18. [26]

    An adaptive distributed predictive control strategy for temperature regulation in a multizone office building,

    F. Lauro, L. Longobardi, and S. Panzieri, “An adaptive distributed predictive control strategy for temperature regulation in a multizone office building,” in2014 IEEE international workshop on intelligent energy systems (IWIES), pp. 32–37, IEEE, 2014

    2014

  19. [27]

    Experimental data-driven model predictive control of a hospital hvac system during regular use,

    E. T. Maddalena, S. A. Mueller, R. M. dos Santos, C. Salzmann, and C. N. Jones, “Experimental data-driven model predictive control of a hospital hvac system during regular use,”Energy and Buildings, vol. 271, p. 112316, 2022

    2022

  20. [28]

    Input convex neural networks for building mpc,

    F. B ¨unning, A. Schalbetter, A. Aboudonia, M. H. de Badyn, P. Heer, and J. Lygeros, “Input convex neural networks for building mpc,” in Learning for Dynamics and Control, pp. 251–262, PMLR, 2021

    2021

  21. [29]

    Experi- mental demonstration of data predictive control for energy optimization and thermal comfort in buildings,

    F. B ¨unning, B. Huber, P. Heer, A. Aboudonia, and J. Lygeros, “Experi- mental demonstration of data predictive control for energy optimization and thermal comfort in buildings,”Energy and Buildings, vol. 211, p. 109792, 2020

    2020

  22. [30]

    Lessons learned from data-driven building control experiments: Contrasting gaus- sian process-based mpc, bilevel deepc, and deep reinforcement learning,

    L. Di Natale, Y . Lian, E. T. Maddalena, J. Shi, and C. N. Jones, “Lessons learned from data-driven building control experiments: Contrasting gaus- sian process-based mpc, bilevel deepc, and deep reinforcement learning,” in2022 IEEE 61st conference on decision and control (CDC...

    2022

  23. [31]

    Physics-informed linear regression is competitive with two machine learning methods in residential building mpc,

    F. B ¨unning, B. Huber, A. Schalbetter, A. Aboudonia, M. H. de Badyn, P. Heer, R. S. Smith, and J. Lygeros, “Physics-informed linear regression is competitive with two machine learning methods in residential building mpc,”Applied Energy, vol. 310, p. 118491, 2022

    2022

  24. [32]

    Provably safe and robust learning-based model predictive control,

    A. Aswani, H. Gonzalez, S. S. Sastry, and C. Tomlin, “Provably safe and robust learning-based model predictive control,”Automatica, vol. 49, no. 5, pp. 1216–1226, 2013

    2013

  25. [33]

    Energy optimal control of thermal comfort in trams,

    R. N. Hofstadter, J. Amaya, and M. Kozek, “Energy optimal control of thermal comfort in trams,”Applied Thermal Engineering, vol. 143, pp. 812–821, 2018

    2018

  26. [34]

    Advanced sliding mode control of heating and ventilation unit in a light rail vehicle,

    A. Shah, D. Huang, T. Huang, and N. Qin, “Advanced sliding mode control of heating and ventilation unit in a light rail vehicle,”Studies in Informatics and Control, vol. 30, no. 3, pp. 121–130, 2021

    2021

  27. [35]

    Neural network based control charting for multiple stream processes with an application to hvac systems in passenger railway vehicles,

    A. Lepore, B. Palumbo, and G. Sposito, “Neural network based control charting for multiple stream processes with an application to hvac systems in passenger railway vehicles,”Applied Stochastic Models in Business and Industry, vol. 38, no. 5, pp. 862–883, 2022

    2022

  28. [36]

    Energy efficiency and indoor thermal comfort of railway carriages: Development of an innovative passenger-centric-control framework for hvac systems,

    A. Buonomano, C. Forzano, G. Giuzio, A. Palombo, and G. Russo, “Energy efficiency and indoor thermal comfort of railway carriages: Development of an innovative passenger-centric-control framework for hvac systems,”Energy, vol. 307, p. 132440, 2024

    2024

  29. [37]

    Design of model-based optimizing control scheme for an air-conditioning system,

    M. Kumar and I. Kar, “Design of model-based optimizing control scheme for an air-conditioning system,”HVAC&R Research, vol. 16, no. 5, pp. 565–597, 2010

    2010

  30. [38]

    The transient predictor,

    K. Moffat, F. D ¨orfler, and A. Chiuso, “The transient predictor,” in2024 IEEE 63rd Conference on Decision and Control (CDC), pp. 1871–1876, IEEE, 2024

    2024

  31. [39]

    Gasser,Machbarkeitsstudie zu den dezentralen Aspekten des pr¨adiktiven thermischen Lastmanagements

    J. Gasser,Machbarkeitsstudie zu den dezentralen Aspekten des pr¨adiktiven thermischen Lastmanagements. feasibility-study by Swiss Federal Railways SBB, chapter 2 ”Modell”, 31.7.2019

    2019

  32. [40]

    Spc: Subspace predictive control,

    W. Favoreel, B. De Moor, and M. Gevers, “Spc: Subspace predictive control,”IFAC Proceedings Volumes, vol. 32, no. 2, pp. 4004–4009, 1999

    1999

  33. [41]

    The bias of subspace-based data-driven predictive control,

    K. Moffat, F. D ¨orfler, and A. Chiuso, “The bias of subspace-based data-driven predictive control,” inThe 2025 IEEE European Control Conference (ECC), IEEE and the European Control Association, 2025

    2025

  34. [42]

    Yalmip : A toolbox for modeling and optimization in matlab,

    J. L ¨ofberg, “Yalmip : A toolbox for modeling and optimization in matlab,” inIn Proceedings of the CACSD Conference, (Taipei, Taiwan), 2004

    2004

  35. [43]

    Physically consistent neural networks for building thermal modeling: theory and analysis,

    L. Di Natale, B. Svetozarevic, P. Heer, and C. N. Jones, “Physically consistent neural networks for building thermal modeling: theory and analysis,”Applied Energy, vol. 325, p. 119806, 2022

    2022

Pith tools

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