Calibrated Simulation

Reddy (2006, p.1) described calibrated simulation as:

“the process of using an existing building simulation computer program and “tuning” or calibrating the various inputs to the program so that observed energy use matches closely with that predicted by the simulation program.”

Once calibrated simulation is achieved, more reliable simulation predictions can be made (ASHRAE, 2009). Calibrated simulation is usually a very useful tool to explore hypothetical, alternative design and operational scenarios and measuring the savings of conservation retrofits to existing buildings (Wang et al., 2012; Aste et al., 2015). However, it is a labour-intensive and time-consuming process that requires a high level of user skill and knowledge in both simulation and practical building operation (ASHRAE, 2009).

In this research, information from the thermal monitoring project regarding on-site recorded weather data, occupancy patterns and the use of MVHR and gas heating systems was used to calibrate the simulation model created using EnergyPlus 8.6 (US Department of Energy, n.d.). EnergyPlus is an open-source, dynamic BPS tool, developed by the Department of Energy (DOE) in the USA. The calibration process was performed using the manual iterative technique (Reddy, 2006; Coakley et al., 2014; Fumo, 2014; Mustafaraj et al., 2014), in which the user of the BPS tool adjusts the input parameters on a trial-and-error basis until the model output matches the recorded data.

3.3.4.1 Empirical Validation of Simulation Results

Measured data from the thermal monitoring project (i.e. zone mean air temperature and heating energy consumption) were plotted against simulation predictions provided from the calibrated

model. The aim was to empirically validate the accuracy of simulation predictions and to understand the sources of uncertainty responsible for any observed divergence. The empirical validation of BPS simulation is a common method used to verify the reliability of simulation predictions (Judkoff & Neymark, 1995; Ryan & Sanquist, 2012). Further details of the use of empirical data for model validation can be found in Section 2.3.4. The divergence between measured data and simulation predictions was quantified using the Mean Biased Error (MBE), the Root Mean Squared Error (RMSE) and the Coefficient of Variation of the Root Mean Squared Error (CV-RMSE) (ASHRAE, 2014; Coakley et al., 2014).

The MBE is a non-dimensional measure of the overall bias in the model, reporting the error between measured and simulated data (Coakley et al., 2014). In this study the MBE is used to indicate whether the model over- or under-predicts the actual performance of the building. However, in the MBE the positive bias compensates for negative bias. Hence, further measures of model error are also required.

The RMSE shows the variability of the data between measured and simulated values. Their difference is calculated and squared for each hour, to overcome the issue of the cancelling effect8. The squared errors are then added and divided by the number of points. A square root of the result is reported as the root mean squared error (RMSE). The RMSE is expressed in the same unit as the base value, allowing to directly correlate the statistical indicator to the actual analysed value (for example temperature is degrees oC).

The Coefficient of Variation of the Root Mean Squared Error (CV-RMSE) is another indicator used to quantify how well the model predicts the actual performance of the building (Granderson & Price, 2013), reported as a percentage. The CV-RMSE is calculated from the

RMSE normalised by the mean of the measured data. It allows to correlate the errors in values that are typically reported in different units (for example kWh and kW).

𝑀𝐵𝐸 (%) =𝛴𝑖=1𝑁 (𝑚𝑖−𝑠𝑖) 𝑚̅ (Eq.8) 𝑅𝑀𝑆𝐸 = √∑𝑁𝑖=1(𝑚𝑖−𝑠𝑖)2 𝑁 (Eq.9) 𝐶𝑉 − 𝑅𝑀𝑆𝐸 (%) =√(𝛴𝑖=1 𝑁 _(𝑚 𝑖−𝑠𝑖)2/𝑁 𝑚̅ (Eq.10) Where,

𝑀𝐵𝐸 is the mean biased error

𝑅𝑀𝑆𝐸 is the root mean squared error

𝐶𝑉 − 𝑅𝑀𝑆𝐸 is the coefficient of variation of the root mean squared error

𝑚_𝑖 and 𝑠_𝑖 are the respective measured and simulated data points for each model instance time step

𝑁 is the number of data points

𝑚̅ is the average of the measured data points This method was used to fulfil Objective No 3.

3.3.4.2

Uncertainty and Sensitivity Analyses

It is generally accepted that there is a high level of uncertainty and sensitivity associated with current BPS methods and tools, which can lead to a lack of confidence in building simulation (Irving, 1982; Macdonald and Strachan, 2001; Hopfe, 2009; Berkeley, Haves and Kolderup, 2014).

In this project probabilistic simulation was performed using Monte Carlo-based global uncertainty and sensitivity analysis (UA/SA) (see Section 2.4.2). The aim was to investigate the robustness of ICF construction method and to determine the sensitivity of ICF simulation

predictions to uncertain input data regarding the material properties of the wall (also known as physical uncertainties). Physical uncertainties refer to the physical properties of the wall materials; thickness (d), thermal conductivity (λ), density (ρ), specific heat capacity (c). Latin Hypercube Sampling (LHS) method was employed as a sampling method to generate sampled variables desirable for the UA (Helton & Davis, 2003; Saltelli et al., 2004; Hopfe, 2009) using SimLab9 2.2.1 (SimLab, no date). The LHS is a probabilistic sampling procedure that incorporates features of both random and stratified sampling (Helton & Davis, 2003). A weight is associated with each sampled element for the estimation of integrals. It is easier to implement than stratified sampling, yet achieves a good coverage of the sample space of the selected elements (Saltelli et al., 2004). The use of LHS method was selected because it increases the sampling performance by increasing the sample uniformity in the hyperspace.

Morris’s method was employed to generate the sampled variables for the SA (Campolongo et al., 2007). Two sampling files were created. In the first one, a normal distribution was assumed for all physical properties under investigation. Each input parameter was assigned a mean (μ) based on the actual construction details from the building case study and a standard deviation (σ) based on information from literature (MacDonald, 2002; Hopfe, 2009). In the second sampling file, a uniform distribution was assumed. For each input parameter the same mean (μ) value was assigned, as before, with a constant ±20% range of variability. A total of 2430 simulations were performed in JEPlus10 (JEPlus, n.d.).

9 _{SimLab is a free software designed by the EU Science Hub, used for Monte Carlo-based UA and SA. It is} composed of three modules: a) Statistical pre-processor (to generate the sample for the UA/SA), b) Model execution module and c) Statistical pro-processor (to perform UA/SA) (SimLab, n.d.).

The process was undertaken in three main steps: 1. Pre-processing

2. Simulation 3. Post-processing

The tools and methods used for each of these steps are shown in Figure 3.3.

Figure 3.3 Three steps within the uncertainty and sensitivity analyses.

UA and SA were employed to fulfil Objective No4. Further information about the uncertainty and sensitivity analyses can be found in Section 4.5 and in the paper in Appendix D.

3.3.4.3

“What-if” Analyses

BPS is often associated with the term virtual laboratory used to conduct virtual experiments to assess the performance of hypothetical, alternative design and operation scenarios and to find quantifiable answers to “what-if” design questions (Attia et al., 2012; Loonen et al., 2014; Clarke & Hensen, 2015). In that respect, calibrated simulation was used to compare the thermal and energy performance of the ICF building case study with that of two hypothetical buildings. The two new building cases were identical to the ICF building (in terms of design, footprint, construction, occupancy, HVAC systems), yet they had different wall constructions

representing a high thermal mass and a low thermal mass building. The ICF simulation model was used as a basecase. Two further simulation models were created to investigate what the energy consumption, internal thermal conditions and dynamic fabric performance of the building would be if the level of thermal mass in the walls was different. The “what-if” comparative method was used to fulfil Objective No5 of the research.