Learning-based Predictive Control Approach for Real-time (PDF)

Virginia Commonwealth University Virginia Commonwealth University VCU Scholars Compass VCU Scholars Compass Theses and Dissertations Graduate School 2021 Learning-based Predictive Control Approach for Real-time Learning-based Predictive Control Approach for Real-time Management of Cyber-physical Systems Management of Cyber-physical Systems Roja Eini Virginia Commonwealth University Follow this and additional works at: https://scholarscompass.vcu.edu/etd Part of the Controls and Control Theory Commons © The Author Downloaded from Downloaded from https://scholarscompass.vcu.edu/etd/6733 This Dissertation is brought to you for free and open access by the Graduate School at VCU Scholars Compass. It has been accepted for inclusion in Theses and Dissertations by an authorized administrator of VCU Scholars Compass. For more information, please contact [email protected].  Learning-based Predictive Control Approach for Real-time Management of Cyber-physical Systems By Roja Eini A Dissertation Submitted to the Faculty of Virginia Commonwealth University in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy in Electrical and Computer Engineering in the Department of Electrical and Computer Engineering Richmond, Virginia August 2021  Copyright by Roja Eini 2021  Learning-based Predictive Control Approach for Real-time Management of Cyber-physical Systems By Roja Eini Approved: Sherif Abdelwahed (Major Professor) Carl Elks (Committee Member) Yanxiao Zhao (Committee Member) Changqing Luo (Committee Member) Milos Manic (Committee Member)  Name: Roja Eini Date of Degree: August , 2021 Institution: Virginia Commonwealth University Major Field: Electrical and Computer Engineering Major Professor: Dr. Sherif Abdelwahed Title of Study: Learning-based Predictive Control Approach for Real-time Management of Cyber-physical Systems Pages of Study: 128 Candidate for Degree of Doctor of Philosophy Cyber-physical systems (CPSs) are composed of heterogeneous, and networked hard- ware and software components tightly integrated with physical elements [72]. Large-scale CPSs are composed of complex components, subject to uncertainties [89], as though their design and development is a challenging task. Achieving reliability and real-time adap- tation to changing environments are some of the challenges involved in large-scale CPSs development [51]. Addressing these challenges requires deep insights into control theory and machine learning. This research presents a learning-based control approach for CPSs management, considering their requirements, specifications, and constraints. Model-based control approaches, such as model predictive control (MPC), are proven to be efficient in the management of CPSs [26]. MPC is a control technique that uses a prediction model to estimate future dynamics of the system and generate an optimal control sequence over a prediction horizon. The main benefit of MPC in CPSs management comes from its ability to take the predictions of system’s environmental conditions and disturbances  into account [26]. In this dissertation, centralized and distributed MPC strategies are designed for the management of CPSs. They are implemented for the thermal management of a CPS case study, smart building. The control goals are optimizing system efficiency (lower thermal power consumption in the building), and improving users’ convenience (maintaining desired indoor thermal conditions in the building). Model-based control strategies are advantageous in the management of CPSs due to their ability to provide system robustness and stability. The performance of a model-based controller strongly depends on the accuracy of the model as a representation of the system dynamics [26]. Accurate modeling of large-scale CPSs is difficult (due to the existence of unmodeled dynamics and uncertainties in the modeling process); therefore, model- based control approach is not practical for these systems [6]. By incorporating machine learning with model-based control strategies, we can address CPS modeling challenges while preserving the advantages of model-based control methods. In this dissertation, a learning-based modeling strategy incorporated with a model-based control approach is proposed to manage energy usage and maintain thermal, visual, and olfactory performance in buildings. Neural networks (NNs) are used to learn the building’s performance criteria, occupant-related parameters, environmental conditions, and operation costs. Control inputs are generated through the model-based predictive controller and based on the learned parameters, to achieve the desired performance. In contrast to the existing building control systems presented in the literature, the proposed management system integrates current and future information of occupants (convenience, comfort, activities), building energy trends, and environment conditions (environmental temperature, humidity,  and light) into the control design. This data is synthesized and evaluated in each instance of decision-making process for managing building subsystems. Thus, the controller can learn complex dynamics and adapt to the changing environment, to achieve optimal performance while satisfying problem constraints. Furthermore, while many prior studies in the filed are focused on optimizing a single aspect of buildings (such as thermal management), and little attention is given to the simultaneous management of all building objectives, our proposed management system is developed considering all buildings’ physical models, environmental conditions, comfort specifications, and occupants’ preferences, and can be applied to various building management applications. The proposed control strategy is implemented to manage indoor conditions and energy consumption in a building, simulated in EnergyPlus software. In addition, for comparison purposes, we designed and simulated a baseline controller for the building under the same conditions. Keywords: Cyber-physical Systems, Smart Building Management System, Model Predictive Control, Machine Learning, Learning-based Control, Model-based Control  DEDICATION I dedicate this work to my best friend and husband, Panos, who has always been there for me not only as a patient spouse, but also as an academic guide and support to lean on. I also dedicate this work to my beloved parents, Esmaeil and Minoo, my dear sister, Sara, and my beloved brother, Reza, who have always empowered me with their unconditional love and support.  ACKNOWLEDGEMENTS First and foremost, I would like to express my deepest appreciation and sincere gratitude to my advisor, Dr. Sherif Abdelwahed for his endless support. I cannot thank him enough for supporting me during the past four years. This dissertation would not have been possible without his valuable comments, encouragement, guidance, and immense knowledge. I would also like to thank my committee members, Dr. Carl Elks, Dr. Yanxiao Zhao, Dr. Changqing Luo, and Dr. Milos Manic for their insightful suggestions and invaluable comments on this work. Lastly, special thanks go to my beloved family and all my friends for their continuous support, love, and blessings.  TABLE OF CONTENTS DEDICATION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ii ACKNOWLEDGEMENTS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iii LIST OF TABLES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii LIST OF FIGURES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viii CHAPTER I. INTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 1.1 Motivations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 1.2 An Overview of Cyber-physical Systems . . . . . . . . . . . . . 4 1.2.1 An overview of building management systems . . . . . . . 5 1.2.2 An overview of the proposed control approach for CPSs management . . . . . . . . . . . . . . . . . . . . . . . . . 6 1.3 Literature Review and Related Works . . . . . . . . . . . . . . . 8 1.3.1 MPC for CPS management . . . . . . . . . . . . . . . . . 8 1.3.2 Model-based Control for CPS management . . . . . . . . 9 1.3.3 Learning-based Control for CPS management . . . . . . . 11 1.3.4 Incorporating Learning with Model-based Control for CPS management . . . . . . . . . . . . . . . . . . . . . . . . . 13 1.4 Summary of Contributions . . . . . . . . . . . . . . . . . . . . . 15 1.5 Dissertation Organization . . . . . . . . . . . . . . . . . . . . . . 18 II. REQUIREMENT SPECIFICATION FOR CYBER-PHYSICAL SYSTEMS 19 2.1 Building Components’ Models . . . . . . . . . . . . . . . . . . . 20 2.1.1 Thermal models . . . . . . . . . . . . . . . . . . . . . . . 21 2.1.2 Humidity models . . . . . . . . . . . . . . . . . . . . . . 24 2.1.3 Occupant behavior models . . . . . . . . . . . . . . . . . 26 2.2 Building Comfort Specifications . . . . . . . . . . . . . . . . . . 29 2.2.1 Thermal comfort . . . . . . . . . . . . . . . . . . . . . . 30 2.2.2 Visual comfort . . . . . . . . . . . . . . . . . . . . . . . . 32  2.2.3 Auditory comfort . . . . . . . . . . . . . . . . . . . . . . 35 2.2.4 Olfactory comfort . . . . . . . . . . . . . . . . . . . . . . 37 2.2.5 Hygienic comfort . . . . . . . . . . . . . . . . . . . . . . 40 2.3 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42 III. DISTRIBUTED AND CENTRALIZED MODEL PREDICTIVE CON- TROL FOR CYBER-PHYSICAL SYSTEMS . . . . . . . . . . . . . . . 43 3.1 Model Predictive Control . . . . . . . . . . . . . . . . . . . . . . 43 3.2 Centralized Model Predictive Control . . . . . . . . . . . . . . . 44 3.3 Distributed Model Predictive Control . . . . . . . . . . . . . . . 47 3.4 Stability Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . 52 3.5 Centralized and Distributed MPC on CPS Case Study . . . . . . . 53 3.5.1 Model definition . . . . . . . . . . . . . . . . . . . . . . . 54 3.5.2 Centralized MPC . . . . . . . . . . . . . . . . . . . . . . 56 3.5.3 Distributed MPC . . . . . . . . . . . . . . . . . . . . . . 58 3.5.4 Simulation results . . . . . . . . . . . . . . . . . . . . . . 63 3.5.5 Practical implementation . . . . . . . . . . . . . . . . . . 66 3.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71 IV. LEARNING-BASED MODEL PREDICTIVE CONTROL FOR CYBER- PHYSICAL SYSTEMS . . . . . . . . . . . . . . . . . . . . . . . . . . . 73 4.1 Learning-based Prediction . . . . . . . . . . . . . . . . . . . . . 74 4.2 Model-based Control Incorporated with Learning . . . . . . . . . 76 4.3 Learning-based MPC for Management of Case Study I . . . . . . 80 4.3.1 Case study I model definition . . . . . . . . . . . . . . . . 80 4.3.2 Learning-based MPC on case study I . . . . . . . . . . . . 80 4.3.3 Simulation results of learning-based MPC on case study I . 83 4.4 Learning-based MPC for Management of Case Study II . . . . . . 88 4.4.1 Case study II model definition . . . . . . . . . . . . . . . 88 4.4.2 Learning-based MPC on case study II . . . . . . . . . . . 89 4.4.3 Simulation results of learning-based MPC on case study II 94 4.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106 V. CONCLUSIONS AND FUTURE RESEARCH . . . . . . . . . . . . . . 109 5.1 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109 5.2 Future Research . . . . . . . . . . . . . . . . . . . . . . . . . . . 114 VI. PUBLICATIONS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117  REFERENCES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119  LIST OF TABLES 2.1 Thermal model parameters . . . . . . . . . . . . . . . . . . . . . . . . . . 22 2.2 Comfort glare index (DGP and DGI) values . . . . . . . . . . . . . . . . . 32 2.3 Comfort luminance threshold levels . . . . . . . . . . . . . . . . . . . . . 33 2.4 Acoustic comfort parameters . . . . . . . . . . . . . . . . . . . . . . . . . 38 2.5 Air quality index levels . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39 2.6 Impacts of excessive Carbon Dioxide on the residents’ body . . . . . . . . 39 2.7 Parameters of IAQDT standard . . . . . . . . . . . . . . . . . . . . . . . . 42 3.1 Thermal model numerical values . . . . . . . . . . . . . . . . . . . . . . . 56 3.2 Numerical characteristics of the state and control signals of the two rooms using centralized and distributed MPC . . . . . . . . . . . . . . . . . . . . 65 4.1 Building materials description . . . . . . . . . . . . . . . . . . . . . . . . 80 4.2 Simulation results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86 4.3 PID parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91 4.4 Performance comparison of baseline and proposed control methods . . . . 105  LIST OF FIGURES 1.1 An overview of the proposed control structure for CPS management . . . . 7 2.1 Electro-thermal circuit model of a single-zone building [15] . . . . . . . . 23 2.2 Electro-thermal circuit model of a multi-zone building [15] . . . . . . . . 23 2.3 Likelihood of a window open considering the indoor temperature [58] . . . 29 2.4 DGP and DGI glare indexes versus the percentage of residents disturbed [114] 34 3.1 Structure of MPC [26] . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44 3.2 Distributed MPC on N interacted subsystems . . . . . . . . . . . . . . . . 47 3.3 Distributed MPC algorithm flowchart . . . . . . . . . . . . . . . . . . . . 51 3.4 Six-room model plan . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55 3.5 Six rooms’ temperature using centralized MPC . . . . . . . . . . . . . . . 64 3.6 Six rooms’ temperature using distributed MPC . . . . . . . . . . . . . . . 65 3.7 Control signal 5 using centralized and distributed MPC . . . . . . . . . . . 66 3.8 The smart home’s CAD plan . . . . . . . . . . . . . . . . . . . . . . . . . 67 3.9 The position of all the sensors and actuators in each floor (top view) . . . . 67 3.10 Picture of one room including its actuators and sensors . . . . . . . . . . . 68 3.11 Picture of sensors, actuators, sources, and control boards used in the smart building . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69 3.12 First-floor temperature trajectory . . . . . . . . . . . . . . . . . . . . . . . 70 3.13 First-floor humidity trajectory . . . . . . . . . . . . . . . . . . . . . . . . 70 3.14 First-floor actuator input and control signal . . . . . . . . . . . . . . . . . 71 4.1 An integrated model-based control and data analytics approach for buildings management . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78 4.2 four-zone building CAD model . . . . . . . . . . . . . . . . . . . . . . . 81 4.3 Learning-based model predictive control (MPC) for thermal management of buildings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82 4.4 NARX neural network model . . . . . . . . . . . . . . . . . . . . . . . . 84 4.5 Neural network output response versus targets . . . . . . . . . . . . . . . . 85 4.6 Regression and performance trajectories of datasets . . . . . . . . . . . . . 86 4.7 Identified model outputs versus real outputs, and the identification error . . 87 4.8 Power consumption and zone 1 temperature using learning-based MPC . . 87 4.9 Power consumption and zone 1 temperature using conventional MPC . . . 88 4.10 Proposed learning-based building control system . . . . . . . . . . . . . . 93 4.11 General block diagram of the simulations . . . . . . . . . . . . . . . . . . 96 4.12 Learned temperature versus targets . . . . . . . . . . . . . . . . . . . . . . 97 4.13 Learned clothing versus targets . . . . . . . . . . . . . . . . . . . . . . . . 97  4.14 Learned PMV thermal comfort index versus targets . . . . . . . . . . . . . 98 4.15 Learned illumination versus targets . . . . . . . . . . . . . . . . . . . . . 99 4.16 Learned glare versus targets . . . . . . . . . . . . . . . . . . . . . . . . . 99 4.17 Learned PPD visual comfort index versus targets . . . . . . . . . . . . . . 100 4.18 Learned CO2 concentration versus targets . . . . . . . . . . . . . . . . . . 100 4.19 Learned PPD olfactory comfort index versus targets . . . . . . . . . . . . 101 4.20 Thermal properties using the proposed control strategy . . . . . . . . . . . 102 4.21 Thermal properties using PID control . . . . . . . . . . . . . . . . . . . . 103 4.22 Visual properties using the proposed control strategy . . . . . . . . . . . . 104 4.23 Visual properties using PID control . . . . . . . . . . . . . . . . . . . . . 104 4.24 Olfactory properties using the proposed control strategy . . . . . . . . . . 106 4.25 Olfactory properties using PID control . . . . . . . . . . . . . . . . . . . . 106