Development of Calibrated Operational Models of Existing ... · Development of Calibrated Operational Models of Existing Buildings for Real-Time Decision Support ... Static Model

CIBSE Technical Symposium, Edinburgh, UK 14-15 April 2016

Page 1 of 14

Development of Calibrated Operational Models of Existing Buildings for Real-Time Decision Support and Performance Optimisation

DANIEL COAKLEY1, 2, GORDON AIRD1, STEPHEN EARLE1, BIRTHE KLEBOW1 AND CATHERINE CONAGHAN1 1 INTEGRATED ENVIRONMENTAL SOLUTIONS LIMITED (IES), GLASGOW, UNITED KINGDOM 2 INFORMATICS RESEARCH UNIT FOR SUSTAINABLE ENGINEERING (IRUSE), NUI GALWAY, IRELAND CORRESPONDING AUTHOR: [email protected]

Abstract

Building simulation tools are commonly used in design for performance appraisal and optimisation. However, numerous studies have found that actual building performance often deviates significantly from simulation predictions. This paper proposes a detailed framework to produce calibrated operational models, which can support operational decision-making, and real-time control optimisation. The approach centres around a three-tier calibration process: Tier 1 focuses on Building-level (Demand-side) variables (e.g. occupancy, equipment, infiltration). Tier 2 focuses on system-level (HVAC) model components (e.g. heating / cooling coil capacities). In this phase, we use detailed building data combined with genetic optimisation techniques to calibrate relevant input parameters. In the case where system performance modelling is not necessary, we use free-form profiles (i.e. measured building data) to supplement these model components. Once system-level noise has been eliminated, in Tier 3 we calibrate the remaining plant-level parameters (e.g. central plant, electricity consumption, etc.). The approach is supported by two novel developments: (1) Free-form profiles: These are actual historic trends from existing building controllers, which are used to supplement model components where appropriate; (2) Genetic Optimisation algorithms are utilised to efficiently navigate the solution space to reduce discrepancies between the model and actual system performance. The proposed calibration approach builds upon prior research efforts to standardise the calibration process using evidence-based model development, combined with sensitivity and uncertainty analysis. Keywords energy simulation, calibration, optimisation, operation

1 Introduction

Up to 90% of the buildings’ life cycle carbon emissions occur during their operational phase, mainly as consequence of the HVAC, lighting and appliances’ energy use (1). Building simulation tools are traditionally used during the design and retrofit stage of buildings (2–4) to enable calculation of building energy demand, occupant thermal comfort, daylighting, etc. While there have been numerous published studies (5–7), of the use of these tools in the operational phase for optimised control and decision support, it is still not part of the mainstream. This is partly due to the disconnect between simulation tools and actual building control systems, as well as a lack of automated strategies for continuous calibration.

mailto:[email protected]


Page 2 of 14

In addition, published research in control for buildings typically use very simple building physics models or “grey/black” box models (8,9). The simple physics models suffer from a lack of modelling capability, which limits their accuracy and scenario prediction capability (5,10,11). In addition, they are often restricted to isolated building systems not addressing the integrated nature of buildings and their systems in operation (11). Building simulation models are usually developed at the design and commissioning stage, but normally they are not later modified again to the actual conditions of the building, unless a new architectural project or retrofitting process is to be accomplished, thus leading to discrepancies between design and actual performance (12–14). As buildings are subject to significant uncertainties and changes during their lifecycle such as weather cycles, facility usage, occupancy, equipment degradation, to name a few, the initial building models might differ from the current building scenario and therefore they might not be valid anymore for assessing the actual situation of the building.

From a review of the existing literature in this area, it is clear that there is an opportunity to improve building control efficiency through the use of more advanced forecasting and performance prediction methods. Detailed physics-based models can play a crucial role in this regard, due to their ability to simulate complex system behaviours and alternative control strategies. However, for this to occur, we first need to bridge the gap between the performance of these predictive models and real systems, through the use of calibration, and continuous re-calibration methods. In addition, we need to clearly outline how such models can interact with existing building controls to support near-real-time decision-making. Energy IN TIME, an EU funded research project, aims to tackle these issues, and develop novel practical approaches for implementation of an integrated simulation-based building controller.

1.1 Energy in Time Project Energy IN TIME project aims to improve existing building control techniques,

developing an integrated control & operation approach, that will combine modelling techniques and simulation software; and new control techniques to obtain automatic generation of optimal operational plans for buildings tailored to the actual requirements. This approach will help negate system inefficiencies and therefore contribute to significantly improving building energy efficiency and comfort. The concept that lies behind Energy IN TIME approach is the use of more accurate building simulation models able to represent buildings’ complexity and uncertainties for building control and operation, while generating forecasted information for use in these models. This will involve the following key developments:

• Development of more advanced calibration techniques through automated processing which will reduce the margin of error to +/-5%, hence improving the accuracy of calibrated models so that they can be used for real-time building control. This will be achieved through the integration of building models (e.g. IES VE, EnergyPlus, TRNsys, IDA ICE, etc.) with the calibration technique taking into account sub-hourly data (opposed to daily or monthly figures).

• Creation of automated calibrated models integrating the energy audit procedure with building models leading to a reduction of the effort and time required in creating a calibrated building model. This further contributes to


Page 3 of 14

the reduction of the margin of error within +/-5% and will be achieved through the integration of the data into the models using an automated import function, developed as part of EiT solution.

• Continuous calibration after the commissioning stage. This will allow for changes to building use with respect to the occupants and adaptation of the buildings environment.

The overall objective of Energy IN TIME is to develop a Smart Energy Simulation Based Control method to reduce the energy consumption and energy bills in the operational stage of existing non-residential buildings. In this paper, we focus specifically on the approach being developed for improved initial model calibration, and subsequent automated continuous calibration. The calibrated model forms the basis for simulation-based control (Refer to Figure 1). However, the model relies on a number of elements from the System or execution level, and provides information at the operational and decision-making level. These components are described in brief below in order to provide context for the purpose of model calibration:

• Building: the physical building / structure under consideration;

• BEMS: Building energy management system, which acts at the data acquisition and control point for the building;

• Faults: In existing buildings, it is assumed that there may be a number of faults present in current operation, which may be classifies as system, sensor or actuator faults. These must be considered as part of the model development in order to avoid compensating for faults during simulation calibration;

• Simulation-based control: the purpose of the model during the building operational phase is to provide performance prediction to enable better control of buildings (i.e. pro-active rather than re-active);

• Fault detection and Diagnosis: detection of system faults by comparing simulation prediction with actual operation;

• Optimal Operational Plan: Continuously updated building operational plan, based on model predictions and forecast data (e.g. energy costs, user behaviour and requirements, weather).

• Decision Support Tool: this is the overall EiT platform which presents all of the information back to the building stakeholder (e.g. facility manager, owner) at a time-scale suitable for decision making.


Page 4 of 14

Figure 1: Energy IN TIME project overview

2 Methodology In this paper we propose an approach based on the best available approaches in

recent scientific literature (15,12), as well as building on recent advances in building energy simulation software. The approach is split in to the following three phases (see Figure 2):

• Stage 1: Model Development and Calibration; • Stage 2: Model Re-calibration; • Stage 3: Control Optimisation.

The Model development and calibration phase is based around existing peer-reviewed publications in the field of detailed simulation calibration using real measured data (12,16–19). Accurate representations of the buildings’ active and passive parts (including entire buildings, systems, and even external spaces) will be developed. The innovation here is that they will be virtual operational models of the buildings with the ability to be tightly calibrated and reflect building situation at any point in time, enabling a realistic representation and simulation of the buildings scenarios. These will be possible thanks to a continuous updating of the models with information obtained from the building monitoring: operational energy loads and data from meters and sensors, such as temperature, relative humidity, CO2 etc., which may be imported directly in to the models in the form of profiles. Subsequently, models will go through an automatic self-calibration. This continuous process will allow models to evolve in parallel with the buildings. The use of these models will allow obtaining representations of the HVAC, lighting and other building systems behaviour with a smaller margin of error. This approach will make possible the prediction of building’s demand and operation conditions providing low deviation with respect to the actual conditions. Additionally, the integration of information about forecasts on occupancy, weather and building loads into these models will permit


Page 5 of 14

obtaining predictions closer to the reality. The outputs of the simulation will lead to the design of highly efficient operational plans (e.g. HVAC schedules and workloads, lighting schedules, etc.) tailored to both buildings and users’ needs.

In this section, we will describe each step in further detail. It should be noted that where the methodology refers to specific software packages below, these are not meant to outline an exhaustive list of options. For our study, we are using IES-VE combined with a range of pre- and post-processing modules, some of which are inextricable linked to the VE software suite. However, the same analytical approach may be applied regardless of choice of modelling software, and the methodology has been developed with this in mind.


Page 6 of 14

Static Model Parameters

Model Profiles <FFP>

Building Operational Data

<SCAN>Base Model <VE>

Sensitivity Analysis

<PB+Python>

Update Model

Performance Criteria Met

NO

Calibrated Base Model <VE>

YES

Re-calibrated Operational Model

Performance Criteria Met?

YES

Automatic re-calibration of Input Profile <Optimise>

NO

Model Variant 1 Model Variant 2 Model Variant 3

Scenario Modelling

Optimal Control DSS

Model Variant 2

Figure 2: Energy in Time Calibration Overview


Page 7 of 14

2.1 Stage 1: Model Development and Calibration In this phase we develop a Base Model of our building or pilot area, using

available historic performance data about the building (static parameters and dynamic profiles). The energy model is developed in two phases, comprising passive and active model components:

• The first stage model will include all the building components that compose the so-called “passive” building parts, including: building envelope, façade, claddings, etc.;

• In the second stage, the active systems of the building are incorporated, that within this context, will be all of those model elements producing and/or consuming energy as well as those distributing energy (e.g. boilers, chillers, HVAC equipment etc.).

The integration of both active-passive models within the simulation environment

produces the base model. This is the initial best guess model of the actual building and system, which shall later be refined during calibration. A base model is similar to a design model that it is created during building design phase (cf. Figure 3). However, as the building considered is already in operation phase and doesn’t need to be constructed anymore, the consultant shouldn’t only consider building design information but also include measured and metered data wherever available.

Figure 3: Base Model Definition If a design model is available, then that will be used as the starting point for the

base model. If there is not existing building energy model, then a standard model development process is used (18–20), involved building audits and collection of relevant data, which can be categorised into two main types:

• Static Values: These are values which should be fixed in the building model, for example, the geometry, constructions, orientation, etc.;

• Dynamic Values: These are non-fixed profile variables which may be dependent on external influences (e.g. weather, occupancy).


Page 8 of 14

Static values are typically attained through on-site inspection or measurement. Where no information is available, information is taken from typical design or compliance guideline values. Dynamic values may be defined in two ways:

• Schedules: Using standard schedule-based profiles within the modelling environment. This is the standard way of defining profiles for building energy models (e.g. lighting, equipment, occupancy, infiltration etc.). However, due to the high-level nature of their definition, it is often difficult to attain an appropriate balance between accuracy and development time.

• Free-form Profiles (FFP): An alternative approach that may be used in the modelling environment uses real data, acquired directly from the building management system, provides a utility for creating model-compatible profiles, based on real data (see Figure 4). The advantage of this approach is that it reduces the time needed to create manual profile definitions for various parameters, and helps to improve the accuracy of the simulation. The caveat is that there is a danger that the model becomes ‘over-specified’, in that the profile used is particular to the time-period under study, and may therefore be less representative of a typical period.

Figure 4: Example of free-form profile (FFP) in red above Once the base model has been defined, Parameter analysis is carried out in

preparation for the actual calibration process (see Figure 5). Its aims are to reduce model complexity to a dimension that can be handled by the available calibration tools, to select dynamic parameters to be monitored and used as input for the operational model and to determine static parameters which values shall be optimised during the calibration process. Sensitivity and uncertainty analysis are methods to

• Understand which model input parameters and variables mainly influence simulation results;

• Identify dependencies of variables, and; • Identify the influence of uncertainties of variable/parameter estimates on the

simulation results.

The importance of parameter analysis for successful model calibration shouldn’t be underestimated.


Page 9 of 14

Figure 5: Model input parameter analysis

The final calibration step can be split up into two sub-steps: • Calibration of static parameters (e.g. geometry, u-values, heater capacities.)

Normally, static parameters don’t change significantly with time. If static parameters change, this often is an indication for a change of building behaviour (e.g. blockage of filters of HVAC systems, …);

• Calibration of dynamic parameters (e.g. weather data, equipment, ventilation, occupancy,). Dynamic parameters cannot be cannot be estimated once as they continuously change and are ideally fed into the model in form of continuously updated profiles.

A process of sensitivity analysis is used to guide the model update and calibration process until model performance criteria are met. The process follows a three-tier structure (see Figure 6), starting at Tier 3 (Building Level) focusing on the calibration of the main demand-side parameters (e.g. zone infiltration, occupancy, equipment, lighting, fabric etc.). Once these parameters have been calibrated to satisfactory performance, the process moves to system (HVAC) calibration, and finally whole-building supply side calibration (e.g. boiler and chiller performance curves and efficiencies). Parameters which are included in the sensitivity analysis and calibration process, at each tier, may be characterised as follows:

• Model design variables: These are parameters that will influence choices during the design (or retrofit) process, but which are not appropriate for change during existing building calibration (e.g. Geometry, Orientation, Constructions, HVAC equipment specifications);

• Generic model variables: These are parameters that will always be of interest during existing building model calibration, and are difficult (or


Page 10 of 14

impossible) to measure, and/or exhibit high degrees of uncertainty (e.g. air infiltration ach, lighting / equipment loads, material conductivity);

• Operation and control scenarios: These are controllable parameters which may be optimised to improve a buildings operational efficiency (e.g. heating/cooling set-points, flow rates, operation schedules);

Figure 6: Three-tier initial model update and calibration process

In the case of static model parameters, these may be assigned best-guess estimates along with appropriate upper and lower bounds. The calibration process involves semi-automated parameter calibration using a parameter batch simulation process to identify the optimal value for individual model parameters (see Figure 7).

Figure 7: Parametric batch simulation for parameter value identification


Page 11 of 14

In the case of dynamic free-form profiles, these may be disaggregated into typical time-varying profiles using a process of wavelet analysis. This will allow for increased flexibility for profile calibration, as well as offering a solution to the problem of over-fitting / over-specification.

The main steps are summarised in Figure 8. While the approach will not automatically calibrate the model entirely, it should provide the following information which will speed up the manual calibration process:

• Level of influence of chosen model parameters (Sensitivity); • Optimum parameter ranges and distributions (Clustering of optimum

solutions);

Figure 8: Calibration Process Summary

2.2 Stage 2: Model Re-calibration As the model will be used during building operation, it is necessary to regularly

assess performance criteria and re-calibrate the model if performance drift occurs. In this phase, uncertain model profiles (e.g. occupancy, infiltration) will be adjusted automatically using an optimisation function. This is known as the Calibrated Operational Model and may be used to make reliable predictions for ongoing building operation and control. At this stage, the concept of disaggregated free-form profiles is used to its full advantage in order to re-evaluate uncertain model profiles to determine the optimal fit to achieve necessary calibration performance criteria. This stage will focus specifically on the Tier 3 model variables, as it is expected that this is where the majority of the building performance uncertainty is introduced, as they are influenced by dynamic externalities such as occupant behaviour and weather.

Model performance is re-assessed at defined intervals (e.g. comparison against validation data set on a daily basis, or prior to decision-making) in order to determine the accuracy and confidence around model prediction on which control or retrofit


Page 12 of 14

decisions are to be based. If the performance fall outside the acceptable range specified at the outset, then automatic recalibration is employed to bring the model back in line with reality. It should be noted that we do not proposed re-adjusting static variables at this point, as these should now be defined to a satisfactory level of confidence, assuming the initial calibration was carried out with sufficient rigour using all available data with which to justify parameter value changes.

2.3 Stage 3: Control Optimisation In the final phase, the calibrated operational model is used to simulate various control scenarios, as well as physical equipment changes (Model Variants) in order to help the building operator understand the impact of proposed energy efficiency measures or control scenarios. The phase relies on the use of short-term forecasts for externalities (>3-day prediction horizon) such as energy prices, weather conditions and occupancy. 3 Future Work The tool will be validated in four European buildings of different uses:

- Test Site 1: Airport in Faro, Portugal - Test Site 2: Office and Test Labs in Bucharest, Romania - Test Site 3: Commercial and Office in Helsinki, Finland - Test Site 4: Hotel in Levi-Lapland, Finland

The validation will be completed against real data

A further publication will discuss the results of this process, and the accuracy of simulation results.

4 Conclusions

In this paper we have outlined a proposed approach for calibrating energy simulation models for use during the operational phase of the building life-cycle. It is proposed that these models may be used to simulate various control scenarios and energy efficiency measures, in order to underpin a decision support system (DSS) for the building operator. The approach follows a three-tier calibration approach, which aims to eliminate noise and discrepancies at the lowest levels (i.e. zone and system component level), before attempting to tackle whole building model performance


Page 13 of 14

evaluation (i.e. at plant level). The approach builds on existing best practices for model calibration, using evidence-based model updates combines with sensitivity analysis to guide the entire process. We have also introduced the concept of utilising real-building data (free-form profiles) from the building management system, where appropriate, to reduce noise, reduce development time and support overall model calibration.

References 1. Ramesh T, Prakash R, Shukla KK. Life cycle energy analysis of buildings: An

overview. Energy Build. 2010;42(10):1592–600. 2. Petersen S, Svendsen S. Method and simulation program informed decisions

in the early stages of building design. Energy Build. 2010 Jul;42(7):1113–9. 3. Augenbroe G. Trends in building simulation. Build Environ. 2002;37(8-9):891–

902. 4. Clarke J a. Energy simulation in building design. Second. 2001. 5. Coffey B, Haghighat F, Morofsky E, Kutrowski E. A software framework for

model predictive control with GenOpt. Energy Build. 2010;42(7):1084–92. 6. Mahdavi A, Orehounig K, Pröglhöf C. A Simulation-supported Control Scheme

For Natural Ventilation In Buildings. Eleventh International IBPSA Conference. Sydney, Australia; 2009. p. 783–8.

7. Liu M, Claridge DE. Use of calibrated HVAC system models to optimize system operation. J Sol energy Eng. American Society of Mechanical Engineers; 1998;120(2):131–8.

8. Afram A, Janabi-Sharifi F. Theory and applications of HVAC control systems - A review of model predictive control (MPC). Build Environ. 2014;72:343–55.

9. Rossiter J a. Model- Based Predictive Control - A practical approach. 2003. 337 p.

10. Touretzky CR, Baldea M. Nonlinear model reduction and model predictive control of residential buildings with energy recovery. J Process Control. Elsevier Ltd; 2014;24(6):723–39.

11. Neto AH, Fiorelli FAS. Comparison between detailed model simulation and artificial neural network for forecasting building energy consumption. Energy Build. 2008 Jan;40(12):2169–76.

12. Coakley D, Raftery P, Keane M. A review of methods to match building energy simulation models to measured data. Renew Sustain Energy Rev. 2014 Sep;37(September):123–41.

13. Demanuele C, Tweddell T, Davies M. Bridging the gap between predicted and actual energy performance in schools. World renewable energy Congress XI. 2010. p. 1–6.

14. Menezes AC, Cripps A, Bouchlaghem D, Buswell R. Predicted vs. actual energy performance of non-domestic buildings: Using post-occupancy evaluation data to reduce the performance gap. Appl Energy. Elsevier Ltd; 2012;97:355–64.

15. Reddy TA. Literature Review on Calibration of Building Energy Simulation Programs: Uses, Problems, Procedures, Uncertainty, and Tools. ASHRAE Trans. 2006 Jan;112(1):226–40.

16. Coakley D, Raftery P, Molloy P. Calibration of Whole Building Energy Simulation Models: Detailed Case Study of a Naturally Ventilated Building


Page 14 of 14

Using Hourly Measured Data. In: Wright J, Cook M, editors. Building Simulation and Optimization. Loughborough, UK; 2012. p. 57–64.

17. Maile T, Bazjanac V, Fischer M. A method to compare simulated and measured data to assess building energy performance. Build Environ. Elsevier Ltd; 2012;56:241–51.

18. Raftery P, Keane M, Costa A. Calibrating whole building energy models: Detailed case study using hourly measured data. Energy Build. 2011 Dec;43(12):3666–79.

19. Raftery P, Keane M, O’Donnell J, O’Donnell J. Calibrating whole building energy models: An evidence-based methodology. Energy Build. 2011 Sep;43(9):2356–64.

20. Coakley D, Raftery P, Molloy P, White G. Calibration of a Detailed BES Model Using an Analytical Optimisation Approach. Proceedings of the 13th International IBPSA Conference. Sydney, Australia; 2011.

5 Acknowledgements

The authors wish to acknowledge support provided by the European Union under the 7th Framework Programme (FP7) for the project Energy IN TIME EeB.NMP.2013-4 (Grant Agreement no. 608981)

http://www.sciencedirect.com/science/article/pii/S0926580515000850#gts0010

Documents

Development of Calibrated Operational Models of Existing ... · Development of Calibrated Operational Models of Existing Buildings for Real-Time Decision Support ... Static Model