Pith. sign in

REVIEW 2 major objections 2 minor 37 references

A Data-Driven Methodology for Scalable Distributed MPC in Heterogeneous Building Aggregation: From Systematic Feature Selection to Convex Optimization

T0 review · 2 major / 2 minor · reviewed 2026-06-28 · grok-4.3

Pith's one-line read A data-driven framework enables distributed MPC for large heterogeneous building aggregations that matches centralized economic performance while scaling to real-time operation.

desk verdict The paper stitches together MPC-aware feature selection, a custom convex transformer, and ADMM into a working pipeline for large building aggregations, but the abstract leaves the actual performance margins and convexity trade-offs unquantified. read the letter →

arxiv 2605.30763 v1 pith:AMEZQ3DH submitted 2026-05-29 eess.SY cs.SY

classification eess.SYcs.SY
keywords distributedMPCbuildingaggregationdemandresponseinput-convextransformerADMMfeatureselectionconvexoptimizationthermalcomfort
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

The paper develops a complete methodology to coordinate demand response across many different buildings using model predictive control. It first applies systematic feature selection tuned to multi-step forecasting needs, then uses a specialized transformer to model building dynamics while keeping the problem convex, and finally solves the coupled optimization in a distributed way with Tracking ADMM. A reader would care because centralized MPC becomes impossible to run in real time once the number of buildings grows large, leaving energy costs and indoor comfort unmanaged under time-varying electricity prices.

What carries the argument

The Input-Convex Encoder-Only Transformer (IC-EoT) that models building dynamics while guaranteeing convexity of the resulting optimization problem, together with the Tracking Alternating Direction Method of Multipliers (ADMM) that solves the constraint-coupled problem in a fully distributed manner.

What would settle it

A co-simulation trial in which the distributed controller's total energy cost for an aggregation of 50 or more buildings deviates by more than a few percent from the centralized optimum or in which its per-step solve time exceeds the real-time control interval.

Watch

Extended reading notes

Core claim

The paper claims that a framework combining MPC-aware feature selection, an Input-Convex Encoder-Only Transformer for dynamics modeling, and Tracking ADMM for distributed solution of the constraint-coupled problem delivers near-identical economic optimality and superior thermal comfort compared to a theoretical centralized controller, while providing the computational scalability required for real-time control of large aggregations of consumer and prosumer buildings, as shown in EnergyPlus co-simulations under a dynamic time-of-use tariff.

Load-bearing premise

The Input-Convex Encoder-Only Transformer accurately captures the thermal dynamics of heterogeneous buildings while ensuring the formulated optimization problem remains convex.

Editorial extensions

If this is right

  • The distributed approach can control aggregations whose size makes centralized MPC computationally infeasible.
  • Economic performance stays nearly identical to the centralized optimum while thermal comfort improves.
  • The method handles mixed consumer and prosumer buildings under dynamic time-of-use pricing.
  • MPC-aware feature selection produces multi-step forecasts robust enough for the subsequent optimization step.

Reading between the lines

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

  • The same combination of convex modeling and distributed ADMM could apply to other large-scale networked systems with coupled constraints, such as district heating or microgrid coordination.
  • Deployment would lower the computing hardware needed at a central operator, shifting effort to local building controllers.
  • Testing the framework on real measured data rather than EnergyPlus simulations would reveal whether model mismatch affects the observed performance gap.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 2 minor

Summary. The paper proposes a data-driven framework for scalable distributed MPC in heterogeneous building aggregations for demand response. It combines an MPC-aware feature selection method to support multi-step forecasting, a novel Input-Convex Encoder-Only Transformer (IC-EoT) to model building dynamics while guaranteeing convexity of the resulting optimization problem, and the Tracking ADMM algorithm to solve the constraint-coupled problem in a fully distributed fashion. High-fidelity co-simulation validation with EnergyPlus models of consumer and prosumer buildings under a dynamic TOU tariff is reported to show near-identical economic optimality and superior thermal comfort relative to a centralized controller, together with computational scalability that renders the centralized approach infeasible for large aggregations.

Significance. If the quantitative validation results and convexity guarantee hold under detailed scrutiny, the work would offer a practical route to real-time distributed MPC for large-scale building demand response, directly addressing the computational intractability of centralized formulations while preserving performance. The systematic MPC-aware feature selection and the IC-EoT architecture represent potentially reusable contributions for convex data-driven modeling in control applications.

major comments (2)
  1. [Abstract] Abstract (validation paragraph): the central performance claims of near-identical economic optimality, superior thermal comfort, and exceptional scalability are stated without any quantitative metrics, error bars, ablation results, or comparison tables. This absence prevents assessment of whether the reported advantages are statistically meaningful or sensitive to the IC-EoT modeling assumptions.
  2. [Framework components] Framework components (IC-EoT description): the assertion that the Input-Convex Encoder-Only Transformer guarantees convexity of the subsequent MPC optimization problem is load-bearing for the entire pipeline, yet no explicit architectural constraints, loss-function modifications, or proof sketch are referenced that would establish this property for arbitrary building dynamics.
minor comments (2)
  1. Define all acronyms at first use (e.g., ADMM, CCP, TOU, DR) and ensure consistent notation between the abstract and the main text.
  2. Clarify the precise interface between the MPC-aware feature selection step and the IC-EoT training objective so that readers can reproduce the multi-step prediction robustness claim.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive comments, which help clarify the presentation of our validation results and the technical foundations of the IC-EoT. We respond to each major comment below and indicate the corresponding revisions.

read point-by-point responses
  1. Referee: [Abstract] Abstract (validation paragraph): the central performance claims of near-identical economic optimality, superior thermal comfort, and exceptional scalability are stated without any quantitative metrics, error bars, ablation results, or comparison tables. This absence prevents assessment of whether the reported advantages are statistically meaningful or sensitive to the IC-EoT modeling assumptions.

    Authors: We agree that the abstract would benefit from quantitative support for the performance claims. In the revised manuscript we will update the validation paragraph to include the specific metrics, error bars, and references to ablation studies and comparison tables already present in the results section. This will enable readers to assess statistical meaningfulness and sensitivity to modeling assumptions directly from the abstract. revision: yes

  2. Referee: [Framework components] Framework components (IC-EoT description): the assertion that the Input-Convex Encoder-Only Transformer guarantees convexity of the subsequent MPC optimization problem is load-bearing for the entire pipeline, yet no explicit architectural constraints, loss-function modifications, or proof sketch are referenced that would establish this property for arbitrary building dynamics.

    Authors: The convexity guarantee rests on the input-convex architecture of the IC-EoT. We will revise the manuscript to make the architectural constraints, loss-function modifications, and proof sketch more explicit in the main text (with the full derivation remaining in the appendix). The proof applies to any dynamics that can be represented by the trained input-convex model; we will add a clarifying sentence on the scope with respect to arbitrary building dynamics. revision: yes

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity detected

full rationale

The paper's pipeline consists of MPC-aware feature selection, the proposed IC-EoT model asserted to produce convex dynamics, and standard ADMM for distributed CCP solving, followed by empirical co-simulation validation against a centralized baseline. No derivation step reduces a claimed prediction or optimality result to a fitted parameter or self-citation by construction; the convexity property is presented as an architectural guarantee of the novel encoder, the performance metrics are reported from external high-fidelity simulation, and no load-bearing uniqueness theorem or ansatz is imported from prior self-work. The central claims therefore remain independent of the inputs they are evaluated against.

Assumptions & free parameters 0 free parameters · 0 assumptions · 1 invented entities

Abstract-only review; no explicit free parameters, axioms, or invented entities beyond the named IC-EoT model can be extracted. The convexity guarantee is treated as a modeling choice rather than a derived property.

invented entities (1)
  • Input-Convex Encoder-Only Transformer (IC-EoT)
    purpose: Model building dynamics while ensuring the MPC optimization remains convex
    Introduced as novel component to guarantee convexity; no independent evidence of convexity preservation provided in abstract.

how reviews work

0 comments
Cite this review

Pith. "Pith review of A Data-Driven Methodology for Scalable Distributed MPC in Heterogeneous Building Aggregation: From Systematic Feature Selection to Convex Optimization." pith.science (2026). https://pith.science/paper/AMEZQ3DH

@misc{pith2026260530763,
  author       = {Pith},
  title        = {Pith review of: A Data-Driven Methodology for Scalable Distributed MPC in Heterogeneous Building Aggregation: From Systematic Feature Selection to Convex Optimization},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/AMEZQ3DH}},
  note         = {Machine review of arXiv:2605.30763}
}
read the original abstract

Coordinating large-scale, heterogeneous building aggregations for demand response (DR) is impeded by a dual challenge: the computational intractability of centralized Model Predictive Control (MPC) and the inadequacy of conventional feature selection methods, which fail to address the error-compounding nature of multi-step forecasting required by MPC. This paper proposes a comprehensive, data-driven framework that first employs a systematic, MPC-aware feature selection methodology to ensure robust multi-step prediction, then models the complex building dynamics using a novel Input-Convex Encoder-Only Transformer (IC-EoT) to guarantee a convex optimization problem, and finally solves the resulting constraint-coupled problem (CCP) in a fully distributed manner using the Tracking Alternating Direction Method of Multipliers (ADMM) algorithm. The framework is validated in a high-fidelity co-simulation environment, controlling a heterogeneous aggregation of consumer and prosumer buildings based on the EnergyPlus under a dynamic time-of-use (TOU) tariff. Results demonstrate that the proposed distributed approach achieves near-identical economic optimality and superior thermal comfort compared to a theoretical centralized controller, while exhibiting exceptional computational scalability that overcomes the real-time infeasibility of the centralized approach for large aggregations.

Figures

Figures reproduced from arXiv: 2605.30763 by the authors.

Figure 1
Figure 1. The proposed three-step systematic feature selection procedure, adapted for learning-based MPC. The process begins with a raw dataset, which is refined [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. The co-simulation framework for the IC-EoT embedded, learning-based distributed MPC. The framework is driven by external inputs: a Weather File feeds [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Mean slover time per control step vs. aggregation size for the T [PITH_FULL_IMAGE:figures/full_fig_p011_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

37 extracted references · 2 canonical work pages

  1. [1]

    europa.eu/topics/energy-efficiency/ energy-efficient-buildings/ energy-performance-buildings-directive_en, accessed: 2023-10-22 (2023)

    European Commission, Energy performance of buildings directive,https://energy.ec. europa.eu/topics/energy-efficiency/ energy-efficient-buildings/ energy-performance-buildings-directive_en, accessed: 2023-10-22 (2023)

  2. [2]

    S.Hu,D.Yan,S.Guo,Y.Cui,B.Dong,Asurveyonenergy consumptionandenergyusagebehaviorofhouseholdsand residential building in urban china, Energy and Buildings 148 (2017) 366–378

  3. [3]

    K. Xu, Z. Zhi, R. Zhao, K. Jiang, End-to-End Differen- tiable Predictive Control with Guaranteed Constraint Sat- isfaction and feasibility for Building Demand Response, arXiv preprint (2025)

  4. [4]

    B. P. Center, Annual energy outlook 2021, Energy Infor- mation Administration, Washington, DC (2021)

  5. [5]

    J. Ma, J. Qin, T. Salsbury, P. Xu, Demand reduction in buildingenergysystemsbasedoneconomicmodelpredic- tive control, Chemical Engineering Science 67 (1) (2012) 92–100

  6. [6]

    D. T. Vedullapalli, R. Hadidi, B. Schroeder, Combined hvac and battery scheduling for demand response in a building, IEEE Transactions on Industry Applications 55 (6) (2019) 7008–7014

  7. [7]

    Drgoňa, J

    J. Drgoňa, J. Arroyo, I. C. Figueroa, D. Blum, K. Arendt, D.Kim,E.P.Ollé,J.Oravec,M.Wetter,D.L.Vrabie,etal., All you need to know about model predictive control for buildings,AnnualReviewsinControl50(2020)190–232

  8. [8]

    Moriyasu, T

    R. Moriyasu, T. Ikeda, S. Kawaguchi, K. Kashima, Struc- tured hammerstein-wiener model learning for model pre- dictive control, IEEE Control Systems Letters 6 (2021) 397–402

Show all 37 references
  1. [9]

    González-Vidal, A

    A. González-Vidal, A. P. Ramallo-González, F. Terroso- Saenz,A.Skarmeta,Datadrivenmodelingforenergycon- sumption prediction in smart buildings, in: 2017 IEEE International Conference on Big Data (Big Data), IEEE, 2017, pp. 4562–4569

  2. [10]

    Azuatalam, W.-L

    D. Azuatalam, W.-L. Lee, F. de Nijs, A. Liebman, Re- inforcement learning for whole-building hvac control and demand response, Energy and AI 2 (2020) 100020

  3. [11]

    Zhang, J

    L. Zhang, J. Wen, A systematic feature selection proce- dure for short-term data-driven building energy forecast- ingmodeldevelopment,EnergyandBuildings183(2019) 428–442

  4. [12]

    Leung, C

    M. Leung, C. Norman, L. L. Lai, T. T. Chow, The use of occupancy space electrical power demand in building cooling load prediction, Energy and Buildings 55 (2012) 151–163

  5. [13]

    Krarti, Energy audit of building systems: an engineer- ing approach, CRC press, 2020

    M. Krarti, Energy audit of building systems: an engineer- ing approach, CRC press, 2020

  6. [14]

    S. B. Taieb, G. Bontempi, A. F. Atiya, A. Sorjamaa, A review and comparison of strategies for multi-step ahead time series forecasting based on the nn5 forecasting com- petition, Expert systems with applications 39 (8) (2012) 7067–7083

  7. [15]

    Diekerhof, F

    M. Diekerhof, F. Peterssen, A. Monti, Hierarchical dis- tributedrobustoptimizationfordemandresponseservices, IEEETransactionsonSmartGrid9(6)(2017)6018–6029

  8. [16]

    Q. Li, Y. Liao, K. Wu, L. Zhang, J. Lin, M. Chen, J. M. Guerrero,D.Abbott,Parallelanddistributedoptimization method with constraint decomposition for energy man- agement of microgrids, IEEE Transactions on Smart Grid 12 (6) (2021) 4627–4640

  9. [17]

    Falsone, I

    A. Falsone, I. Notarnicola, G. Notarstefano, M. Prandini, Tracking-admm for distributed constraint-coupled opti- mization, Automatica 117 (2020) 108962. 12

  10. [18]

    C.Pathiravasam, G.K.Venayagamoorthy, Distributedde- mandresponsemanagementforavirtuallyconnectedcom- munity with solar power, IEEE Access 10 (2022) 8350– 8362

  11. [19]

    Myroshnychenko, S

    I. Myroshnychenko, S. Podosynnikov, D. Halynskyi, M.Ushkalov,O.Chuhai,etal.,Regulatorybarriersforen- trepreneurship and start-ups in renewable energy: Biblio- metric analysis, SocioEconomic Challenges 8 (3) (2024) 181–210

  12. [20]

    Khafiso, C

    T. Khafiso, C. Aigbavboa, S. A. Adekunle, Barriers to theadoptionofenergymanagementsystemsinresidential buildings, Facilities 42 (15/16) (2024) 107–125

  13. [21]

    SURV,Comunidadesinteligentes,eficientesysostenibles, http://www3.uah.es/eneffpilot, accessed: 2022- 05-06 (2022)

  14. [22]

    K.Xu,Z.Zhi,K.Jiang,InputConvexEncoder-OnlyTrans- former for Fast and Gradient-Stable MPC in Building De- mandResponse: AchievingReal-TimeOptimizationunder Practical Time Horizons, arXiv preprint (2025)

  15. [23]

    Sankaranarayanan, R

    P. Sankaranarayanan, R. Rengaswamy, Cdinn–convex dif- ference neural networks, Neurocomputing 495 (2022) 153–168

  16. [24]

    M.-C. Paré, V. Dermardiros, A. Lesage-Landry, Effi- cientdata-drivenmpcfordemandresponseofcommercial buildings, arXiv preprint arXiv:2401.15742 (2024)

  17. [25]

    S. Xing, J. Zhang, S. Li, J. Gao, A bilevel fast-convergent optimizer via high-fidelity convex models: Application on optimal operation of all-parallel heterogeneous chiller- pump systems, Energy and Buildings 314 (2024) 114238

  18. [26]

    B.Amos,L.Xu,J.Z.Kolter,Inputconvexneuralnetworks, in: International conference on machine learning, PMLR, 2017, pp. 146–155

  19. [27]

    H. Zhu, P. Tan, Z. He, C. Zhang, Q. Fang, G. Chen, Non- linearmodelpredictivecontrolofuscboiler-turbinepower units in flexible operations via input convex neural net- work, Energy 255 (2022) 124486

  20. [28]

    Y. Chen, Y. Shi, B. Zhang, Optimal control via neural networks: A convex approach, arXiv preprint arXiv:1805.11835 (2018)

  21. [29]

    Z.Wang,D.Yu,Z.Wu,Real-timemachine-learning-based optimization using input convex long short-term memory network, Applied Energy 377 (2025) 124472

  22. [30]

    Bünning, A

    F. Bünning, A. Schalbetter, A. Aboudonia, M. H. de Ba- dyn, P. Heer, J. Lygeros, Input convex neural networks for building mpc, in: Learning for Dynamics and Control, PMLR, 2021, pp. 251–262

  23. [31]

    Q.Li,Q.Meng,J.Cai,H.Yoshino,A.Mochida,Applying support vector machine to predict hourly cooling load in the building, Applied Energy 86 (10) (2009) 2249–2256

  24. [32]

    Kapetanakis, E

    D.-S. Kapetanakis, E. Mangina, D. P. Finn, Input variable selectionforthermalloadpredictivemodelsofcommercial buildings, Energy and Buildings 137 (2017) 13–26

  25. [33]

    B. He, L. Hou, X. Yuan, On full jacobian decomposition of the augmented lagrangian method for separable con- vex programming, SIAM Journal on Optimization 25 (4) (2015) 2274–2312

  26. [34]

    Falsone, M

    A. Falsone, M. Prandini, Augmented lagrangian tracking for distributed optimization with equality and inequality coupling constraints, Automatica 157 (2023) 111269

  27. [35]

    H. Peng, F. Long, C. Ding, Feature selection based on mutual information criteria of max-dependency, max- relevance, and min-redundancy, IEEE Transactions on pattern analysis and machine intelligence 27 (8) (2005) 1226–1238

  28. [36]

    D. B. Crawley, L. K. Lawrie, F. C. Winkelmann, W. F. Buhl, Y. J. Huang, C. O. Pedersen, R. K. Strand, R. J. Liesen,D.E.Fisher,M.J.Witte,etal.,Energyplus: creat- inganew-generationbuildingenergysimulationprogram, Energy and buildings 33 (4) (2001) 319–331

  29. [37]

    Politi, Energym: A building model library for controller benchmarking, Applied Sciences 11 (8) (2021) 3518

    P.Scharnhorst,B.Schubnel,C.FernandezBandera,J.Sa- lom,P.Taddeo,M.Boegli,T.Gorecki,Y.Stauffer,A.Pep- pas, C. Politi, Energym: A building model library for controller benchmarking, Applied Sciences 11 (8) (2021) 3518. 13

Pith tools

Reviewed June 28, 2026 · model on record in the stance chip above.