Physics-constrained Deep Learning of Multi-zone Building Thermal Dynamics

We present a physics-constrained control-oriented deep learning method for modeling building thermal dynamics. The proposed method is based on the systematic encoding of physics-based prior knowledge into a structured recurrent neural architecture. Specifically, our method incorporates structural priors from traditional physics-based building modeling into the neural network thermal dynamics model structure. Further, we leverage penalty methods to provide inequality constraints, thereby bounding predictions within physically realistic and safe operating ranges. Observing that stable eigenvalues accurately characterize the dissipativeness of the system, we additionally use a constrained matrix parameterization based on the Perron-Frobenius theorem to bound the dominant eigenvalues of the building thermal model parameter matrices. We demonstrate the proposed data-driven modeling approach's effectiveness and physical interpretability on a dataset obtained from a real-world office building with 20 thermal zones. Using only 10 days' measurements for training, we demonstrate generalization over 20 consecutive days, significantly improving the accuracy compared to prior state-of-the-art results reported in the literature.

[1]  Andrew Kusiak,et al.  Modeling and optimization of HVAC systems using a dynamic neural network , 2012 .

[2]  Peter R. Armstrong,et al.  Modeling Environment for Model Predictive Control of Buildings , 2014 .

[3]  Jaume Salom,et al.  State-space models for building control: how deep should you go? , 2020, Journal of Building Performance Simulation.

[4]  Zheng O'Neill,et al.  Comparisons of inverse modeling approaches for predicting building energy performance , 2015 .

[5]  Damien Picard,et al.  Comparison of Model Predictive Control Performance Using Grey-Box and White-Box Controller Models of a Multi-zone Office Building , 2016 .

[6]  Hamid Reza Shaker,et al.  Comparative Analysis of White-, Gray- and Black-box Models for Thermal Simulation of Indoor Environment: Teaching Building Case Study , 2018 .

[7]  Lei Chen,et al.  A neural network-based multi-zone modelling approach for predictive control system design in commercial buildings , 2015 .

[8]  Sirish L. Shah,et al.  MPC relevant identification––tuning the noise model , 2004 .

[9]  Christian Koch,et al.  Building information modelling based building energy modelling: A review , 2019, Applied Energy.

[10]  Dirk Müller,et al.  Automated data-driven modeling of building energy systems via machine learning algorithms , 2019, Energy and Buildings.

[11]  S. Baldi,et al.  Dual estimation: Constructing building energy models from data sampled at low rate , 2016 .

[12]  Yoshua Bengio,et al.  Tackling Climate Change with Machine Learning , 2019, ACM Comput. Surv..

[13]  Jimmy Ba,et al.  Adam: A Method for Stochastic Optimization , 2014, ICLR.

[14]  Henrik Madsen,et al.  Identifying suitable models for the heat dynamics of buildings , 2011 .

[15]  Eldad Haber,et al.  Stable architectures for deep neural networks , 2017, ArXiv.

[16]  Dirk Saelens,et al.  Quality of grey-box models and identified parameters as function of the accuracy of input and observation signals , 2014 .

[17]  Pascal Fua,et al.  Imposing Hard Constraints on Deep Networks: Promises and Limitations , 2017, CVPR 2017.

[18]  David Duvenaud,et al.  Neural Ordinary Differential Equations , 2018, NeurIPS.

[19]  Thomas B. Schön,et al.  Linearly Constrained Neural Networks , 2020, ArXiv.

[20]  Rahul Mangharam,et al.  Data predictive control using regression trees and ensemble learning , 2017, 2017 IEEE 56th Annual Conference on Decision and Control (CDC).

[21]  Razvan Pascanu,et al.  Understanding the exploding gradient problem , 2012, ArXiv.

[22]  Gm. Shafiullah,et al.  Modeling techniques used in building HVAC control systems: A review , 2017 .

[23]  Lieve Helsen,et al.  Identification of multi-zone grey-box building models for use in model predictive control , 2020 .

[24]  Michael Wetter,et al.  ModestPy: An Open-Source Python Tool for Parameter Estimation in Functional Mock-up Units , 2018, Proceedings of The American Modelica Conference 2018, October 9-10, Somberg Conference Center, Cambridge MA, USA.

[25]  Jason Yosinski,et al.  Hamiltonian Neural Networks , 2019, NeurIPS.

[26]  Zhiping Mao,et al.  DeepXDE: A Deep Learning Library for Solving Differential Equations , 2019, AAAI Spring Symposium: MLPS.

[27]  Natalia Gimelshein,et al.  PyTorch: An Imperative Style, High-Performance Deep Learning Library , 2019, NeurIPS.

[28]  Zhipeng Jia,et al.  Constrained Deep Weak Supervision for Histopathology Image Segmentation , 2017, IEEE Transactions on Medical Imaging.

[29]  Roberto Lamberts,et al.  Application and characterization of metamodels based on artificial neural networks for building performance simulation: A systematic review , 2020 .

[30]  Manfred Morari,et al.  Analysis of energy savings potentials for integrated room automation , 2010 .

[31]  Damien Picard,et al.  Approximate model predictive building control via machine learning , 2018 .

[32]  Jing Yuan,et al.  Log-barrier constrained CNNs , 2019, ArXiv.

[33]  Ján Drgoňa,et al.  Constrained Neural Ordinary Differential Equations with Stability Guarantees , 2020, ICLR 2020.

[34]  Philip Haves,et al.  Model predictive control for the operation of building cooling systems , 2010, Proceedings of the 2010 American Control Conference.

[35]  Alberto Sangiovanni-Vincentelli,et al.  Building Efficiency and Sustainability in the Tropics ( SinBerBEST ) Title Online Simultaneous State Estimation and Parameter Adaptation for Building Predictive , 2013 .

[36]  Dario Ambrosini,et al.  Data-driven model predictive control using random forests for building energy optimization and climate control , 2018, Applied Energy.

[37]  P. Schmid,et al.  Dynamic mode decomposition of numerical and experimental data , 2008, Journal of Fluid Mechanics.

[38]  Piljae Im,et al.  Aggregation and Data Driven Identification of Building Thermal Dynamic Model and Unmeasured Disturbance , 2020, Energy and Buildings.

[39]  D. Laurí,et al.  PLS-based model predictive control relevant identification: PLS-PH algorithm , 2010 .

[40]  S. C. Kremer,et al.  Gradient Flow in Recurrent Nets: the Difficulty of Learning Long-Term Dependencies , 2001 .

[41]  Johan Åkesson,et al.  Toolbox for development and validation of grey-box building models for forecasting and control , 2014 .

[42]  Gregor P. Henze,et al.  A model predictive control optimization environment for real-time commercial building application , 2013 .

[43]  Rafael E. Carrillo,et al.  A Hybrid Learning Method for System Identification and Optimal Control , 2020, IEEE Transactions on Neural Networks and Learning Systems.

[44]  Steven L. Brunton,et al.  On dynamic mode decomposition: Theory and applications , 2013, 1312.0041.

[45]  Eleni Mangina,et al.  Data-driven Predictive Control for Unlocking Building Energy Flexibility: A Review , 2020, ArXiv.

[46]  D. Vrabie,et al.  Constrained Physics-Informed Deep Learning for Stable System Identification and Control of Unknown Linear Systems , 2020, ArXiv.

[47]  J. Maciejowski Guaranteed stability with subspace methods , 1995 .

[48]  Trevor Darrell,et al.  Constrained Convolutional Neural Networks for Weakly Supervised Segmentation , 2015, 2015 IEEE International Conference on Computer Vision (ICCV).

[49]  Frauke Oldewurtel,et al.  Experimental analysis of model predictive control for an energy efficient building heating system , 2011 .

[50]  L. H. Hansen,et al.  Modelling the heat dynamics of a building using stochastic differential equations , 2000 .

[51]  Uri Shalit,et al.  Structured Inference Networks for Nonlinear State Space Models , 2016, AAAI.

[52]  Draguna Vrabie,et al.  A Learning Framework for Control-Oriented Modeling of Buildings , 2017, 2017 16th IEEE International Conference on Machine Learning and Applications (ICMLA).

[53]  Í. Ciglera,et al.  Beyond theory : the challenge of implementing Model Predictive Control in buildings Ji ř , 2013 .

[54]  Damien Picard,et al.  Impact of the controller model complexity on model predictive control performance for buildings , 2017 .

[55]  Andrew G. Alleyne,et al.  Decentralized predictive thermal control for buildings , 2014 .

[56]  Michael Wetter,et al.  All you need to know about model predictive control for buildings , 2020, Annu. Rev. Control..

[57]  Lukas Ferkl,et al.  Ceiling radiant cooling: Comparison of ARMAX and subspace identification modelling methods , 2010 .

[58]  Alessia Arteconi,et al.  Performance Assessment of Data-Driven and Physical-Based Models to Predict Building Energy Demand in Model Predictive Controls , 2020 .

[59]  Lieve Helsen,et al.  TACO, an automated toolchain for model predictive control of building systems: implementation and verification , 2018, Journal of Building Performance Simulation.

[60]  Jian Sun,et al.  Deep Residual Learning for Image Recognition , 2015, 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR).

[61]  Alberto Bemporad,et al.  Learning Nonlinear State-Space Models Using Deep Autoencoders , 2018, 2018 IEEE Conference on Decision and Control (CDC).

[62]  Ruben Villegas,et al.  Learning Latent Dynamics for Planning from Pixels , 2018, ICML.

[63]  Thierry S. Nouidui,et al.  Modelica Buildings library , 2014 .

[64]  J. Zico Kolter,et al.  Learning Stable Deep Dynamics Models , 2020, NeurIPS.

[65]  Damien Picard,et al.  Cloud-based implementation of white-box model predictive control for a GEOTABS office building: A field test demonstration , 2020 .

[66]  Kevin Gimpel,et al.  Bridging Nonlinearities and Stochastic Regularizers with Gaussian Error Linear Units , 2016, ArXiv.

[67]  Bart De Moor,et al.  N4SID: Subspace algorithms for the identification of combined deterministic-stochastic systems , 1994, Autom..

[68]  Dirk Saelens,et al.  Implementation and verification of the IDEAS building energy simulation library , 2018 .

[69]  Matthias W. Seeger,et al.  Deep State Space Models for Time Series Forecasting , 2018, NeurIPS.

[70]  Chi Keong Goh,et al.  A Constrained Deep Neural Network for Ordinal Regression , 2018, 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition.

[71]  António E. Ruano,et al.  Prediction of building's temperature using neural networks models , 2006 .

[72]  Mohamed Benbouzid,et al.  Intelligent Systems for Building Energy and Occupant Comfort Optimization: A State of the Art Review and Recommendations , 2018 .

[73]  Eran Treister,et al.  IMEXnet: A Forward Stable Deep Neural Network , 2019, ICML.

[74]  Steve B. Jiang,et al.  Nonlinear Systems Identification Using Deep Dynamic Neural Networks , 2016, ArXiv.

[75]  Jan Peters,et al.  Deep Lagrangian Networks: Using Physics as Model Prior for Deep Learning , 2019, ICLR.

[76]  Farrokh Janabi-Sharifi,et al.  Review of modeling methods for HVAC systems , 2014 .

[77]  Gabriele Pannocchia,et al.  An Open-Source System Identification Package for Multivariable Processes , 2018, 2018 UKACC 12th International Conference on Control (CONTROL).

[78]  Tianbao Yang,et al.  Advancing non-convex and constrained learning: challenges and opportunities , 2019, SIGAI.

[79]  Hao Huang,et al.  Model predictive control for energy-efficient buildings: An airport terminal building study , 2014, 11th IEEE International Conference on Control & Automation (ICCA).

[80]  Frauke Oldewurtel,et al.  Use of partial least squares within the control relevant identification for buildings , 2013 .