Content uploaded by Julian Bischof
Author content
All content in this area was uploaded by Julian Bischof on Oct 20, 2022
Content may be subject to copyright.
D
EV E L O P M E N T O F A
P
Y T H O N
-
BA S E D S I M PL I F I E D H O U R LY B U I L D I N G
M O D E L F O R N O N
-
D O M E S T I C BU I L D I N G S T OC K
O
P E R AT I O N A L E N E R G Y
S I M U LA TI O N S
Julian Bischof
1,2
, Simon Knoll
3
, Aidan Duffy
2
1
Institute for Housing and Environment (Institut Wohnen und Umwelt (IWU)) - Research Institute of
the State of Hesse and the City of Darmstadt, Darmstadt, Germany, E-Mail: j.bischof@iwu.de
2
Dublin Energy Lab and School of Civil and Structural Engineering, Technological University
Dublin, Dublin, Ireland, E-Mail: aidan.duffy@tu-dublin.ie
3
E-Mail: simon.knoll@gmx.net
Abstract
Building stock (BS) energy simulation is an important
tool for exploring a BS’s greenhouse gas mitigation
measures. Simulation presents the challenge of finding
the trade-off between detail and computational
requirements. This paper presents the development
and validation of such a model based on the simplified
hourly method of the ISO 13790, implemented in
Python. The model was validated using 464 sample
buildings from recently collected German non-
domestic building stock data. The validation displayed
an acceptable level of accuracy for the aggregated
building stock with a bias of 19.3 % and 11.2 % for
space heating and electricity use, respectively.
Introduction
The global climate crisis (IPCC, 2021) is pushing the
global transition to a post-fossil fuel age. Buildings
contribute significantly to energy use and greenhouse
gas (GHG) emissions. The operation of residential and
non-residential buildings cause 17 % and 10 % of all
global energy-related GHG-emission respectively
(International Energy Agency, 2021). Accounting for
indirect embodied emissions, the global building stock
(BS) is responsible for approximately 37 % of global
energy-related GHG-emission (International Energy
Agency, 2021). In Germany, the non-residential and
residential building stocks accounted for 14.2 % and
27.4 % of total final energy consumption in 2010,
respectively (IEA-BEEP, 2019). Therefore, the energy
transition in the building sector must play an important
role in climate protection policies.
Building stock models (BSMs) are important
policymaking tools for exploring options to
decarbonise building stocks, (Julian Bischof and
Aidan Duffy, 2022; Röck et al., 2021; Mastrucci et al.,
2017). The development of such models must satisfy
several, often competing, objectives. Policymakers
require these models to be representative, simple to
use and easy to understand. Model developers require
sufficient flexibility to amend and extend the model to
adapt it to new requirements. Users also require
acceptable execution time (Bischof and Duffy, 2021)
and outputs which cover the main building energy end
uses including space heating and cooling, and
appliances. An hourly time step allows the model to
capture transient behaviours such as embedded
generation and grid balancing.
Several building energy models (BEMs) exist which
could be used in the development of a BSM. These can
be divided into either complex or simple models.
Typically, complex models have been developed for
the simulation of single buildings with comprehensive
in- and outputs. They are detailed and accurate due to
the very detailed physical representations employed
which require many input parameters. For this reason,
complex models are problematic to implement in
BSMs since this level of input variable detail is often
not available at a stock level (Malhotra et al., 2021;
Julian Bischof and Aidan Duffy, 2022). Moreover,
their complexity necessitates significant
computational resources for multiple buildings
resulting in long execution times. Common examples
of such complex models are EnergyPlus and INSEL
(Malhotra et al., 2021).
Simple BEMs have been developed to work with
fewer model inputs to approximate energy demand
(simulated energy use) based on the most dominant
energy flows, thereby resulting in faster execution
times (Lim and Zhai, 2017). A further advantage of
such models is that they are easier to understand by
non-experts such as policymakers, thus increasing the
acceptance of the model (Bischof and Duffy, 2021).
Examples include models based on the simplified
hourly method of ISO 13790 or, more recently, ISO
52016 (Malhotra et al., 2021; Julian Bischof and
Aidan Duffy, 2022).
Building stock energy and emissions models which
use physical stock data are typically based on these
relatively simple BEMs. However, the vast majority
of these focus on domestic stocks, with only few
examples for non-domestic stocks. The recently
completed ENOB:DataNWG (see www.datanwg.de)
project involved collecting statistically representative
physical (e.g. energy, building fabric and systems)
data on the German non-domestic building stock.
However, for the reasons outlined above, no readily
available model for its operational building energy
simulation could be identified.
Aim and Methodology
A user requirements survey (Bischof and Duffy, 2021)
and a state-of-the-art review (Julian Bischof and
Aidan Duffy, 2022) have identified the need for a
suitable and feasible energy computer simulation
model for non-domestic building stocks which can be
used to provide knowledge for policymaking. More
specifically, the model should be physics-based,
relatively simple, easy to understand, reliable and easy
to adapt and based on best practice standards such as
the ISO 13790. It should rely on the minimum
required input variables, and these should be generally
available for building stocks. To facilitate maximum
user access, the software should be open source, and
simulation times should be relatively short (less than
one hour) for commonly-available desktop PCs.
Finally, any simulation model developed must be
tested and validated to ensure its reliability.
The development of the simulation tool involved the
following steps:
1. identification of a model scope and most
appropriate methodology;
2. screening suitable existing models as a basis
for development;
3. identification of a suitable data-set for model
simulation and validation;
4. extending the existing model to meet the
required scope;
5. model simulation; and
6. model validation.
These steps are described in more detail below.
Model scope, methodology identification
and model screening (Steps 1 and 2)
A detailed model screening was undertaken
considering 98 models identified in (Julian Bischof
and Aidan Duffy, 2022), six in (Malhotra et al., 2021),
one open-source non-domestic BEM available on
GitHub (Jayathissa, 2020) and the VSA 2.0 model
recently developed by (Bischof, 2021). In total 106
models were considered when selecting the most
appropriate existing model to meet the requirements
outlined above. The selection involved sifting the 106
models based on: suitability for non-domestic
buildings; suitability for archetype/disaggregated
building; the use of a building physics approach;
appropriate output requirements (heating, cooling and
electricity for appliances); confirmed validation or
verification; open-source availability; model
simplicity relative to other models; and free-to-use
software.
Applying these criteria, the ISO 13790 based
RC_BuildingSimulator (Jayathissa, 2020; Jayathissa
et al., 2017) was identified as the most suitable model.
This uses a widely accepted simplified hourly thermal
network model with 5 resistances to the heat flow and
one capacity for internal heat storage (5R1C). An
open-source software version of the model – the
Python-based RC_BuildingSimulator - was selected
for the task of model development. The ISO 13790
model has two key advantages: it is simple, with few
inputs required; and the method has been previously
verified and validated (Maccarini et al., 2021). Its
successor standard (ISO 52016) was not selected due
to is its higher complexity (e.g. requiring highly
disaggregated individual building elements) and more
temperature nodes. While the additional detail
improves model accuracy for individual buildings
(assuming the necessary data are available), for
building stocks, where detailed building element data
are unavailable and must be inferred or interpolated,
the additional model complexity only adds greater
uncertainty, thus reducing simulation accuracy (van
Dijk, 2020). This greater simplicity also means that
the ISO 13790 model would have shorter execution
times (Felsmann et al., 2020; van Dijk, 2020).
Validation data-set (Step 3)
Internationally, there are very limited data available
for simulating non-domestic building stocks. One
exception is the ENOB:DataNWG-Project which has
made statistically representative data available on the
energy-related characteristics and refurbishment rates
of the German non-domestic building stock (further
details available at: www.datanwg.de). While basic
data were collected for larger samples of buildings,
detailed data for 464 buildings were collected using
on-site surveys. This involved measuring energy
consumption, detailed physical building
characteristics and the appliance usage parameters
necessary for energy demand simulations, model
calibration and validation (detailed documentation
and variable descriptions are available at
shorturl.at/bkIU7; building characteristics at
shorturl.at/jtJSY; and measured energy use at
shorturl.at/opGJN). Although the sample was biased
(public bodies, for example, were overrepresented and
transport buildings representing about 1 % of the
German NDBS are not included), this did not affect its
usefulness for model validation. However, for this
reason, no extrapolations to the building stock level
are undertaken here.
Model methodology (Step 4)
In the ISO 13790 5R1C model the internal air
temperature node (θair) is influenced by internal gains
(Φint), solar gains (Φsol) and heating and cooling power
(ΦHC,nd) (see Figure 1). The latter (ΦHC,nd) provides the
energy in- and outflow for conditioning the zone to
stay within certain set temperature set points. The
internal and solar gains act on the internal surface (θs)
and the thermal mass (θm) temperature nodes. θair is
connected to the supply air temperature node (θsup);
this relationship is modelled using the ventilation heat
transfer coefficient (Hve). Heat transport between θs
and θair is modelled by the heat transfer coefficient
(HTC) (Htr,is). θs is also influenced by the external air
temperature (θe) via the HTC for transparent surfaces
and doors (Htr,w) and to the thermal mass temperature
node (θm) via the HTC Htr,ms representing the internal
heat transfer process for the opaque elements (Htr,op).
θm itself is coupled to θe through the external part of
Htr,op the HTC Htr,em. θm is further connected to the
internal thermal storage capacity of the building (Cm).
The details of the calculation of the resistances and
storage capacity are described in the ISO 13790 as
well as the available model documentation (see
shorturl.at/cfsHS), and are therefore not repeated here.
In this paper, a new non-domestic stock model is
proposed. The Dynamic ISO Building Simulator
(DIBS), uses the 5R1C model described above and is
adapted for simulating German non-domestic
buildings. It includes occupancy schedules, appliance
gains and lighting requirements according to the DIN
V 18599-10 and the SIA 2024 for the calculation of
the internal gains Φint through people’s metabolism
and their use of appliances and lighting. The natural
lighting contribution to lighting demand and solar
gains Φsol are estimated using window areas and
directions. A location-based sun position model which
considers direct and diffuse solar radiation is used
based on (Quaschning and Hanitsch, 1995), coupled
with assigned weather data of a typical meteorological
year (TMY) for the nearest available weather station.
A total of 93 weather station TMYs are available in
the model. The choice of sun position and weather
station is based on each building’s postal code.
In addition to its adaptation to a German environment,
the model has been extended to incorporate aspects
missing in the simplified ISO 13790 implementation
of the RC_BuildingSimulator, but which are important
for realistically simulating the demand for space
heating, cooling and appliance electricity use.
Figure 1: ISO 13790 5R1C simple hourly method
network model for one building zone
For example, Φsol is adapted to include shading, and
so reduce transparent surfaces’ solar transmittance
during summertime (1st of April to 1st of October); this
applies in all cases where θe is above 24°C and thereby
above the desired temperature in case of cooling
(θi,c,soll) for θair. This threshold applies to 86 % of the
usage zones defined in the DIN V 18599-10 profiles.
Where these conditions are met, the glass energy
transmittance is reduced to a defined value depending
on the type of window and shading available
The ventilation heat transfer coefficient (Hve) was also
adjusted to take account of occupancy periods so as to
provide a estimate of ventilation heat losses. Here,
occupancy schedules specify the periods of natural
ventilation and mechanical ventilation to meet
minimum air change rates. When unoccupied,
infiltration only is considered. Appliance electricity
use, although not part of the ISO 13790, is estimated
based on the standard values of the DIN V 18599-10.
Lighting demand is also considered during the
occupancy periods. The final energy demand
considers the efficiencies of all typical heating and
cooling systems.
The DIBS model, including detailed documentation
and a data preprocessor that directly translates survey
coded variables and assigns them to DIBS input
variables of on-site inspection data, is available under
an open-source license on Github:
https://github.com/IWUGERMANY/DIBS---
Dynamic-ISO-Building-Simulator.
Inputs
The DIBS uses a total of 44 input variables providing
information on the building’s location, geometry,
usage (including setpoints and gains), installed
systems, their efficiencies and control, internal loads,
physical attributes (e.g. lighting utilisation and
transmittance) of the transparent surfaces, thermal
properties of the building envelope, air change rates
and the thermal capacity of the building. Many of
these inputs are automatically generated in the data
preprocessing module based on on-site inspection
data. The assignment of default input variables is
based on the building usage category (e.g. lighting
load), window-type (e.g. glass light transmittance) and
construction type (e.g. thermal capacity). A full list of
all required input variables is available at
shorturl.at/gmtBZ.
Outputs
The DIBS estimates the energy demand and the final
energy use for space heating and cooling,
differentiated by electricity and other energy carriers.
The electricity demand for lighting and appliance use
is also estimated. Because an engineering model
methodology has been adopted, other intermediate
results (e.g. the solar gains of transparent surfaces) of
interest can also be outputted where desired.
Validation
Space heating and small power/lighting electricity
demand were used to validate the DIBS, since
measured data were available for both of these
variables. Space cooling requirements are included
with electricity since these data are aggregated in the
dataset. To ensure a like-for-like comparison,
measured data are climate- and vacancy-corrected by
employing the approach used for calculating energy
ratings for non-domestic buildings (Worm, 2015) (see
shorturl.at/bkIU7).
Objects representing extreme outliers of over
600 kWh/m2·a for the measured energy use or the
simulated demand are excluded from the validation, as
they are likely to be erroneous.
Results (Steps 5 and 6)
Before the ENOB:dataNWG survey data could be
inputted into the model, it required preprocessing. For
this reason, a data preprocessor was developed. The
subsequent simulation takes about 5 seconds per
building on an average Laptop (Lenovo ThinkPad
L480, Win10 x64, Intel Core i7-8550U CPU @
1.80 GHz 2.00 GHz (8 Cores), 31,9 GB RAM (see
shorturl.at/kAQ57 for details)).
The comparison of the on-site inspection stock (406
buildings with both measured and simulated space
heating demand and 411 buildings for electricity)
shows that on average the simulated space heating
demand is 19.3 % above the measured, while the
electricity demand is 11.2 % greater. Comparing the
results based on the building usage category (listed in
Table 1) a diverse picture unfolds (see Figure 2 and
Table 2). While the OAGB, Trade and TU building
Table 1: Building usage categories included in the
validation data set
categories show a good statistical agreement (< ±21 %
PBias), the model overestimates space heating for the
categories BHRC, CuLe, HC, RaUT and Sch, and,
underestimates PWWO and SF. The RaUT buildings
are overestimated by more than 280 % PBias. The
scatterplot with linear trendlines for the individual
usage categories and overall sample (ALL) is
presented in Figure 3 and shows a general
overestimation for higher values as well as
underestimation for lower values, suggesting a
systematic problem. Similar findings were made, for
example, by (Hörner and Lichtmeß, 2017) regarding
domestic buildings in Luxemburg.
With regard to electricity (see Figure 4), the simulated
and measured electricity demands are similar (< 17 %
|PBias|) for building usage categories CuLe, OAGB
and RaUT, but BHRC, HC and PWWO are
overestimated, while SF, Sch, TU and Trade are
underestimated. Simular to space heating, an
individual building comparison indicates a systematic
overestimation of higher values while at the same time
underestimation of lower energy use. The high values
in the measurements are most likely caused by the
inclusion of electric consumers not considered in the
standard values used for demand simulation (e.g. more
production-related consumers such as server units).
A full summary of validation figures and tables is
available at shorturl.at/zBGV1.
Discussion and Conclusion
The simulation of space heating and electrical energy
requirements using the DIBS model resulted in overall
mean absolute errors of 99,77 and 45,11 kWh/(m2·a)
respectively, showing the potential for model
improvement on an individual building level. These
results were achieved using input variables which are
commonly available in building stock databases.
However, model accuracy was lower for certain
building usage categories (see Table 2). This was
likely the result of unrepresentative category-specific
input zonal occupancy parameters such as temperature
setpoints, lighting attributes (controls and load), air
change rates, internal gains (occupancy and
appliance), occupancy patterns and occupancy
Figure 2: Comparison of simulated space heating
demand (blue boxes) to the measured space heating
consumption (red boxes), by building usage category
Table 2: Statistical analysis of model results for
space heating and their performance against
measurements for the entire Stock (ALL) and
individual usage categories (see Table 1)
Figure 3: Scatterplot of simulated space heating
demand to the measured space heating consumption,
by building usage category
Figure 4: Comparison of simulated electricity
demand (blue boxes) to the measured electricity
consumption (red boxes), by building usage category
intensities. A sensitivity analysis found that the air
change rate for natural ventilation (windows) had the
greatest impact on the space heating demand.
One option to improve the accuracy of the DIBS
building usage category outputs is to use building
usage category average occupancy values, instead of
the values of the most prominent usage zone in the
building as it is done currently. For example in case of
an OAGB building use the occupancy variables should
be based on an area weighted average of: the usage
zones of offices, kitchens, sanitary, storage,
conference and traffic areas instead of applying the
specific parameters for the office use on the entire
building. Further, the default air change rates of
building categories greatly overestimate the space
heating demand and should be adjusted. Additionally,
as there might be more hard-to-identify variables
contributing to systematic errors (e.g. the assigned
average u-values based on building age and
construction type), a second improvement option is to
apply statistical learning techniques to create
calibration factors to “correct” the sum of the
remaining systematic biases.
In addition to these possible improvements, it is
planned to extend DIBS to simulate hot water use.
Furthermore, DIBS will be used in conjunction with
the statistically representative ENOB:dataNWG
interview data-set to simulate the behaviour of the
German non-domestic building stock under different
input assumptions, allowing extrapolation on to the
entrie stock.
Summary
This paper describes the development,
implementation and validation of the DIBS model
which builds on the ISO 13790 5R1C hourly
simplified model. Compared to other stock simulation
models, DIBS employs a relatively small number of
input variables which are typically available in stock
databases. Due to its simplicity, it results in a short
computation time of 5 s per building on an average
laptop. The model mean outputs were 19.3 % and
11.2 % above the measured space heating and
electricity energy use respectively for a sample of the
German non-domestic building stock. Significantly
greater errors were observed for building-use sub-
categories and for individual buildings which will be
addressed in further work.
Acknowledgement
The FlexGeber project, funding code Fk z.:
03EGB0001 and the ENOB:dataNWG project,
funding code Fkz.: 03ET1315, are funded by the
German Federal Ministry for Economic Affairs and
Energy in accordance with the parliamentary
resolution of the German Parliament.
CRediT authorship contribution
Julian Bischof: Conceptualization, Methodology,
Investigation, Programming, Formal analysis,
Visualization, Writing original draft, Writing – review
& editing. Simon Knoll: Programming,
Methodology, Visualisation, Writing – review &
editing. Aidan Duffy: Supervision, Writing – review
& editing.
References
Bischof J. Verbrauchsstrukturanalyse VSA 2.0.
Online: IWU, 2021. 18 p.
Bischof J, Duffy A. Non-domestic building stock
energy and carbon modelling for policy advice –
a user requirements survey. Journal of Physics:
Conference Series 2021; 2021.
Felsmann C, Perschk A, Franke R. Test Of ISO
52016-1 Energy Performance Of Buildings
Calculation Procedure. In: Corrado V, Fabrizio E,
Gasparella A, Patuzzi F, editors. Proceedings of
Building Simulation 2019: 16th Conference of
IBPSA. Building Simulation Conference
proceedings. IBPSA; 2020. p. 3910–3916.
Hörner M, Lichtmeß M. Energy performance of
buildings 2017; 2017: 10.
IEA-BEEP. Buildings Energy Efficiency Policies
Database - Ireland,
https://www.iea.org/beep/Ireland/; 2019 [accessed
May 20, 2019].
International Energy Agency. 2021 GLOBAL
STATUS REPORT FOR BUILDINGS AND
CONSTRUCTION. Nairobi, Kenya, 2021. 105 p.
IPCC. Climate Change 2021 - The Physical Science
Basis (AR6). WMO & UNEP 2021; 2021.
Jayathissa P. RC_BuildingSimulator,
https://github.com/architecture-building-
systems/RC_BuildingSimulator; 2020 [accessed
February 24, 2021].
Jayathissa P, Luzzatto M, Schmidli J, Hofer J, Nagy
Z, Schlueter A. Optimising building net energy
demand with dynamic BIPV shading. Applied
Energy 2017; 202: 726–35.
Julian Bischof, Aidan Duffy. Life-cycle assessment
of non-domestic building stocks: A meta-analysis
of current modelling methods. Renewable and
Sustainable Energy Reviews 2022; 2021(153
(2022) 111743).
Lim H, Zhai ZJ. Review on stochastic modeling
methods for building stock energy prediction.
Building Simulation 2017; 10(5): 607–24.
Maccarini A, Prataviera E, Zarrella A, Afshari A.
Development of a Modelica-based simplified
building model for district energy simulations.
Journal of Physics Conference series 2021; 2021.
Malhotra A, Bischof J, Nichersu A, Häfele K-H,
Exenberger J, Sood D, et al. Information
modelling for urban building energy simulation—
A taxonomic review. Building and Environment
2021: 108552.
Mastrucci A, Marvuglia A, Leopold U, Benetto E.
Life Cycle Assessment of building stocks from
urban to transnational scales. Renewable and
Sustainable Energy Reviews 2017; 74: 316–32.
Quaschning V, Hanitsch R. SHADE
CALCULATIONS IN PHOTOVOLTAIC
SYSTEMS. ISES Solar World Conference 1995;
1995.
Röck M, Baldereschi E, Verellen E, Passer A, Sala S,
Allacker K. Environmental modelling of building
stocks – An integrated review of life cycle-based
assessment models to support EU policy making,
2021. 49 p.
van Dijk D. EN ISO 52016-1: The New International
Standard To Calculate Building Energy Needs for
Heating And Cooling, Internal Temperatures And
Heating And Cooling Load. In: Corrado V,
Fabrizio E, Gasparella A, Patuzzi F, editors.
Proceedings of Building Simulation 2019: 16th
Conference of IBPSA. Building Simulation
Conference proceedings. IBPSA; 2020. p. 4061–
4068.
Worm. Bekanntmachung der Regeln für
Energieverbrauchswerte im
Wohngebäudebestand. BMUB, 2015.
