Elsevier

Applied Energy

Volume 154, 15 September 2015, Pages 829-841
Applied Energy

Model-based and model-free “plug-and-play” building energy efficient control

https://doi.org/10.1016/j.apenergy.2015.05.081Get rights and content

Highlights

  • “Plug-and-play” Building Optimization and Control (BOC) driven by building data.

  • Ability to handle the large-scale and complex nature of the BOC problem.

  • Adaptation to learn the optimal BOC policy when no building model is available.

  • Comparisons with rule-based and advanced BOC strategies.

  • Simulation and real-life experiments in a ten-office building.

Abstract

Considerable research efforts in Building Optimization and Control (BOC) have been directed toward the development of “plug-and-play” BOC systems that can achieve energy efficiency without compromising thermal comfort and without the need of qualified personnel engaged in a tedious and time-consuming manual fine-tuning phase. In this paper, we report on how a recently introduced Parametrized Cognitive Adaptive Optimization – abbreviated as PCAO – can be used toward the design of both model-based and model-free “plug-and-play” BOC systems, with minimum human effort required to accomplish the design. In the model-based case, PCAO assesses the performance of its control strategy via a simulation model of the building dynamics; in the model-free case, PCAO optimizes its control strategy without relying on any model of the building dynamics. Extensive simulation and real-life experiments performed on a 10-office building demonstrate the effectiveness of the PCAO–BOC system in providing significant energy efficiency and improved thermal comfort. The mechanisms embedded within PCAO render it capable of automatically and quickly learning an efficient BOC strategy either in the presence of complex nonlinear simulation models of the building dynamics (model-based) or when no model for the building dynamics is available (model-free). Comparative studies with alternative state-of-the-art BOC systems show the effectiveness of the PCAO–BOC solution.

Introduction

Several studies, both from the US Department of Energy and from the EU Directorate of Energy, have demonstrated the potentialities of Building Optimization and Control (BOC) systems to provide improved energy efficiency in commercial and residential buildings [1], [2]. A BOC system, also referred to as energy management system, is a system of computer-aided tools aiming at monitoring and controlling electrical building loads, mainly heating, ventilation and air conditioning (HVAC) units [3]. Deploying BOC systems is becoming more and more affordable and easy: however, their practical use is still very limited and their energy efficiency potentialities are not fully exploited: currently, the vast majority of BOC-equipped buildings adopts simple rule-based BOC logics with limited energy efficiency capabilities [4], [5]. The main reason for the limited use of more advanced BOC systems that can potentially provide more significant energy efficiency than rule-based solutions is the fact that the design and deployment of advanced BOC systems requires a tedious and time-consuming manual calibration procedure. The need for a manual calibration procedure comes from the fact that most advanced BOC design approaches are based on simplified (typically linear) models for the building dynamics, as the use of more complex and realistic models renders the BOC design practically not possible in most situations. This is the case, for example, of model predictive control-based BOC systems, where the use of linear models makes the optimization task feasibly implementable, while the use of nonlinear building models makes it computationally cumbersome [6], [7], [8], [9], [10], [11]. As a result of using simplified models, the derived advanced BOC systems, when applied in the real-world buildings, might exhibit suboptimal performance which, sometimes, can be poor or even unacceptable [12], [13]: qualified personnel is then extensively engaged in a period of calibration activity that is required to elevate the advanced BOC system to its best achievable performance. Such calibration procedure can be either assisted by Building Energy Performance Simulation (BEPS) models providing elaborate and accurate descriptions of the building dynamics [14], [15], [16], or directly performed on the real building by tuning the BOC system during its implementation. In both cases, the fine-tuning procedure is time-consuming, conducted manually, and mainly relying on expertise and human judgment.

Recently, there has been a considerable research effort toward the development of “plug-and-play” BOC designs which have the potential of overcoming the need for a time-consuming manual fine-tuning procedure. Such “plug-and-play” approaches do not need a simplified model for the building dynamics but, instead, they are “plugged” directly either to an elaborate model for the building dynamics or to the actual building and they accomplish the BOC design automatically. More precisely:

  • Model-based “plug-and-play” BOC designs utilize available BEPS models for assessing the energy performance of the controlled building [17], [18], [19], [20], [21]. A BEPS model of the building is connected to a parameterized BOC system; the significant advantage of model-based designs is that an optimizer updates the parameters of the BOC system by evaluating them via the BEPS model, possibly using realistic conditions and physical constraints that occur during real-life building operations.

  • Model-free “plug-and-play” BOC designs proceed one step further by avoiding the need for a BEPS model of the building [22], [23], [24], [25], [26], [27], [28], [29]. Instead, they are plugged directly into the actual building and they accomplish the control design “on the fly”, i.e., while the building is in operation. Model-free designs possess the significant advantage over model-based ones that they do not inherit inaccuracies of the model: in fact, no matter how elaborate a BEPS model is, the presence of inaccuracies is inevitable and, due to wearing and aging phenomena, inaccuracies become larger as times evolves.

Despite the above-mentioned advantages, the development of efficient model-based and model-free “plug-and-play” BOC systems faces the following challenges:

  • i.

    In the case of model-based BOC designs, the challenge is to employ optimizers that can operate over elaborate building simulation models. Unfortunately, even for buildings of small size (e.g., with 4–5 rooms), the complex interactions between the different building elements might negatively affect the efficiency of most state-of-the-art optimizers. In general, designing an efficient BOC system becomes a hard optimization task as it requires optimizing over hundreds of parameters that are mutually interacting in a nonlinear and complex manner.

  • ii.

    In the case of model-free BOC designs, the challenge is to embed the design with efficient self-tuning mechanisms for adapting in real-time the BOC parameters. In fact, the incorporation of adaptation introduces a learning transient of – sometimes – long duration that may lead to non-acceptable performance. As a matter of fact, the available results in adaptive and self-tuning control have exhibited an efficient performance only in systems of small-scale with linear or linear-like dynamics [30], [31]. However, buildings have highly multivariable nonlinear dynamics.

Recently, we have introduced and analyzed a novel “plug-and-play” control design approach – namely Parametrized Cognitive Adaptive Optimization (PCAO) [32], [33] – which is applicable to both model-based and model-free designs and which successfully addresses the challenges described above. The purpose of this paper is to report on the results of the implementation of PCAO for the design of model-based and model-free “plug-and-play” BOC systems. Simulation and real-life experiments conducted on a 10-office building are reported. The experiments show the capability of PCAO to efficiently cope with a large number of parameters that are mutually interacting in a highly nonlinear manner. Furthermore, the PCAO self-tuning mechanisms are capable of adapting the control system in a safe manner without introducing poor transient performance.

The paper is organized as follows: Section 2 presents the building that served as the test case for running the reported experiments. The PCAO BOC algorithm is presented in Section 3. Sections 4 Simulation experiments, 5 Real-life experiments collect the simulation and real-life experiments performed on the test case. The experiments have been performed as part of a wider effort, the FP7-funded research project AGILE. The interested reader can find more information on [34], [35] or http://www.agile-fp7.eu/. Section 6 summarizes the main findings and concludes the paper.

Section snippets

Test case and BOC objectives

The test case for evaluating the proposed BOC systems is a 10-office building (see Fig. 1) located at the Technical University of Crete in Chania, Greece. All the offices of the test case building are equipped with indoor temperature, humidity and occupancy sensors, and split type HVAC units. The building test case covers a surface of 450 m2: the annual energy consumption of the building for HVAC operation is 130 kW h/m2 (based on energy audits), which means 58,500 kW h every year and on average 160 

PCAO concept and main principles

Many control problems can be formulated as optimization problems aiming at minimizing an objective function which is related to the system performance. In case of BOC system design, the problem is the one of minimizing the total score (1) subject the building dynamics. Contrary to many optimization problems, where the systems dynamics are assumed to be known and described by a state-space analytical model, in the building case such assumptions are rather restrictive. Building models based on

Simulation experiments

In this and in the next section, we provide the results of extensive simulation as well as real-life experiments performed on the test case building (on its EnergyPlus model, and on the real building itself) under different BOC strategies. The model-based and the model-free PCAO BOC system will be hereafter shortened with MB-BOC and MF-BOC, respectively. Please notice that all the data in the plots and in the tables refer to the daily mean energy and discomfort averaged to one office, which

Real-life experiments

Several real-life experiments were conducted in the test case building during summers of 2012 and 2013. For the real-life experiments, the sensor/actuator infrastructure deployed as part of the FP7-funded project PEBBLE [36] was utilized. In all real-life experiments, only the PCAO–BOC system with L = 1 has been tested. Since the interest of the experiments was on reduced energy consumption, only the RB-BOC2 (with 25 °C) was used for comparisons. The main challenges in evaluating real-life

Conclusions

The main conclusions drawn from both the simulation and real-life experiments are:

  • The model-based PCAO BOC design is able to provide efficient solutions to large-scale BOC problems where state-of-the-art optimization-based designs are at stake or totally fail. Different performance measures have been adopted to show consistent improvements in front of weather variability;

  • The model-free PCAO BOC design is able to learn the nearly-optimal BOC strategy “on the fly”, directly from measurements

Acknowledgements

The research leading to these results has been partially funded by the European Commission FP7-ICT-5-3.5, Engineering of Networked Monitoring and Control Systems, under the contract #257806 AGILE – http://www.agile-fp7.eu/ and FP7-ICT-2013.3.4, Advanced computing, Embedded Control Systems #611538 Local4Global – http://www.local4global-fp7.eu/.

References (54)

  • D. Crawley et al.

    Contrasting the capabilities of building energy performance simulation programs

    Build Environ

    (2008)
  • D. Crawley et al.

    EnergyPlus: creating a new-generation building energy simulation program

    Energ Buildings

    (2001)
  • J. Wright et al.

    Optimization of building thermal design and control by multi-criterion genetic algorithm

    Energ Buildings

    (2002)
  • E.H. Mathews et al.

    Improved thermal building management with the aid of integrated dynamic HVAC simulation

    Build Environ

    (2003)
  • R. Velik et al.

    Grid-price-dependent energy management in microgrids using a modified simulated annealing triple-optimizer

    Appl Energy

    (2014)
  • A. Dounis et al.

    Indoor air-quality control by a fuzzy-reasoning machine in naturally ventilated buildings

    Appl Energy

    (1996)
  • J.K.W. Wong et al.

    Intelligent building research: a review

    Autom Constr

    (2005)
  • J. Bai et al.

    A new adaptive PI controller and its application in HVAC systems

    Energy Convers Manage

    (2007)
  • H. Doukas et al.

    Intelligent building energy management system using rule sets

    Build Environ

    (2007)
  • P.M. Ferreira et al.

    Neural networks based predictive control for thermal comfort and energy savings in public buildings

    Energ Buildings

    (2012)
  • G. Kyriakarakos et al.

    Intelligent demand side energy management system for autonomous polygeneration microgrids

    Appl Energy

    (2013)
  • R. Karunakaran et al.

    Energy efficient fuzzy based combined variable refrigerant volume and variable air volume air conditioning system for buildings

    Appl Energy

    (2010)
  • M. Adolph et al.

    Adaptive control strategies for single room heating

    Energ Buildings

    (2014)
  • L. Perez-Lombard et al.

    A review on buildings energy consumption information

    Energ Buildings

    (2008)
  • W. Chung

    Using the fuzzy linear regression method to benchmark the energy efficiency of commercial buildings

    Appl Energy

    (2012)
  • P. de Wilde

    The gap between predicted and measured energy performance of buildings: a framework for investigation

    Autom Constr

    (2014)
  • Roth KW, Westphalen D, Dieckmann J, Hamilton SD, Goetzler W. Energy consumption characteristics of commercial building...
  • Cited by (68)

    • Data requirements and performance evaluation of model predictive control in buildings: A modeling perspective

      2021, Renewable and Sustainable Energy Reviews
      Citation Excerpt :

      Consequently, extracting the relationship between model and control performance involves explicitly designed experiments, which is rare in the past years. Several studies showed that model mismatch could result in more energy consumption [119,132] and/or discomfort [97,139]. To quantify, 10% error led to 5% more energy cost and 100% more comfort violation [28].

    • Deep reinforcement learning for energy management in a microgrid with flexible demand

      2021, Sustainable Energy, Grids and Networks
      Citation Excerpt :

      Model-free or data-driven approaches consist of learning abstract representations of near-optimal control strategies in the microgrid from its operational data. Learning-based methods have been introduced in recent years as an alternative to model-based approaches, as they can reduce the need for an explicit system model, improve the EMS scalability, and reduce the maintenance costs of the EMS [21]. One of the most promising learning-based EMS methods is the reinforcement learning (RL) paradigm [22], in which an agent learns the dynamics of the microgrid by interacting with its components.

    View all citing articles on Scopus
    1

    The first two authors have equally contributed to this work.

    View full text