This article was downloaded by: [Aneta Strzalka]On: 18 August 2011, At: 00:26Publisher: Taylor & FrancisInforma Ltd Registered in England and Wales Registered Number: 1072954 Registeredoffice: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK

HVAC&R ResearchPublication details, including instructions for authors andsubscription information:http://www.tandfonline.com/loi/uhvc20

3D City modeling for urban scale heatingenergy demand forecastingAneta Strzalka a , Jürgen Bogdahn b , Volker Coors b & Ursula Eicker aa Center of Applied Research, Sustainable Energy Technology,University of Applied Sciences Stuttgart, Schellingstrasse 24, 70174,Stuttgart, Germanyb Department of Geomatics, Computer Science and Mathematics,University of Applied Sciences Stuttgart, Schellingstrasse 24, 70174,Stuttgart, Germany

Available online: 17 Aug 2011

To cite this article: Aneta Strzalka, Jürgen Bogdahn, Volker Coors & Ursula Eicker (2011): 3D Citymodeling for urban scale heating energy demand forecasting, HVAC&R Research, 17:4, 526-539

To link to this article: http://dx.doi.org/10.1080/10789669.2011.582920

PLEASE SCROLL DOWN FOR ARTICLE

Full terms and conditions of use: http://www.tandfonline.com/page/terms-and-conditions

This article may be used for research, teaching and private study purposes. Anysubstantial or systematic reproduction, re-distribution, re-selling, loan, sub-licensing,systematic supply or distribution in any form to anyone is expressly forbidden.

The publisher does not give any warranty express or implied or make any representationthat the contents will be complete or accurate or up to date. The accuracy of anyinstructions, formulae and drug doses should be independently verified with primarysources. The publisher shall not be liable for any loss, actions, claims, proceedings,demand or costs or damages whatsoever or howsoever caused arising directly or indirectlyin connection with or arising out of the use of this material.

3D City modeling for urban scale heating energy demandforecasting

Aneta Strzalka,1,∗ Jurgen Bogdahn,2 Volker Coors,2 and Ursula Eicker11Center of Applied Research, Sustainable Energy Technology, University of Applied Sciences Stuttgart,

Schellingstrasse 24, 70174 Stuttgart, Germany2Department of Geomatics, Computer Science and Mathematics, University of Applied Sciences Stuttgart,

Schellingstrasse 24, 70174 Stuttgart, Germany∗Corresponding author e-mail: aneta.strzalka@hft-stuttgart.de

An urban energy management tool was developed, which is able to predict the heating energy demand ofurban districts and analyze strategies for improving building standards. Building models of different Levelsof Detail are investigated and analyzed according to their suitability for forecasting energy demand. Basedon the specific 3D city model, an input file is generated, which can be read by the building simulation model.Special focus is put on a method for modeling the heating energy demand of the buildings with the fewestinput parameters possible, but one which will give reliable forecast results. A simple transmission heat lossmethod and an energy-balance method were tested. In both cases, there was a good correlation betweenthe measured and calculated annual values for a case study area of over 700 buildings in Ostfildern,Germany. The results also show that a 3D city model (with low geometrical detail) can be used for energydemand forecasting on an urban scale.

Introduction

There are many modeling techniques availablethat can be used to model the energy consumption ofan urban area (Swan and Ugursal 2009). City-scaleprediction of the heating energy demand relies onthe accuracy of input data (Level of Detail), whichcan vary significantly. The Level of Detail of the in-put data for simulations depends on data availabil-ity and can strongly influence the type of modelingtechnique that is chosen. The easiest modeling tech-nique for the heating energy demand of an entire cityis based on typification of districts, where the pre-diction of the heating energy demand is based on thedistrict type, its size, and the number of buildings

Received December 15, 2010; accepted March 29, 2011Aneta Strzalka is PhD Student and Researcher. Jurgen Bogdahn is PhD Student and Researcher. Volker Coors is Professor.Ursula Eicker is Professor and Scientific Director.

in it. This method does not consider the buildingsseparately, but the size of the district type in squarekilometers as well as the the age of the buildings(Blesl 2002). A very similar method was introducedby Firth and Lomas (2009), in which the energyprediction is made for each category of dwellingrather than for each individual dwelling in the com-munity in the study. The energy prediction is madefor an average national dwelling, and the results arethen multiplied by the number of dwellings of thistype of building in the community. A more detailedmethod, also described by Blesl (2002), consideredeach building separately, whereby the heated grossarea is estimated by the multiplication of the build-ing ground area with the number of floors. A higher

526

HVAC&R Research, 17(4):526–539, 2011. Copyright C© 2011 American Society of Heating, Refrigerating and Air-Conditioning Engineers,Inc.ISSN: 1078-9669 print / 1938-5587 onlineDOI: 10.1080/10789669.2011.582920

Dow

nloa

ded

by [

Ane

ta S

trza

lka]

at 0

0:26

18

Aug

ust 2

011

HVAC&R RESEARCH 527

precision in determining the building geometry isreached using laser scanning data. Blesl combinedthe obtained building volumes with the building ty-pology in order to estimate thermal parameters, suchas heat transfer coefficients (U-values) (Blesl 2010).To verify the estimated U-values for the appropriatebuilding type, simultaneously taken infrared imagesof the city quarter are used. The data is then used tocalculate the heating energy demand according tothe European Directive (DIN e.V. 2003). Assump-tions are made for the window area ,and the user be-havior is defined with a standard load curve. Heat-uptimes are not taken into account.

A rather new method, called rapid energy mod-eling (Autodesk White Paper 2011), presents a wayto identify buildings that have the greatest potentialfor achieving energy savings with minimal effortand cost. This method is a process of energy de-mand prediction involving very few data and makesuse of the image capture of building exteriors andsimplified simulation. This method uses simple dig-ital pictures of the buildings to extract buildingsexteriors. Hereby, the 2D images are calibrated ina special software in order to create 3D buildingmodels. These 3D CAD files are then converted toa building information model. Finally, the build-ing information models are used as an input for aWeb-based simulation of the whole-building energydemand, including electricity. The main goal was toshow how effectively this method could be used toestimate the energy usage of existing buildings. Tocompare the calculated electricity demand values,the actual energy usage of the buildings from theutility bills was used. The deviations between themeasured and calculated values by analysis of sev-eral buildings was at most 20%.

The calculation models used to predict the heat-ing energy demand vary from very easy static en-ergy balance, or degree-day-based calculations, tomore complex dynamic building simulation mod-els. The simple degree-day method estimates theheating energy demand using the inputs of the dif-ference between a constant room temperature andambient temperature and the parameters of the over-all heat transfer coefficient (Jaffal et al. 2009). Asmentioned in Mavrogianni et al. (2009), most of thebuilding simulation models that are available on themarket require a large amount of data input, whichcan lead to the problem of being able to acquire suf-ficient data to model the heating energy demand of

an entire city. Therefore, an important aspect of thisresearch was to find a solution that makes it pos-sible to forecast the heating energy demand of thebuildings while using the least amount of input pa-rameters possible, while still giving reliable results.This also involves a study of the detail of availablemodels and how reliable the simulation results ofthese models are.

The weak point in the design of most of the avail-able city-scale heating energy demand predictionmethods is the lack of the validation process. Ac-cording to Mathews et al. (1997), the success ofthe model development process depends on its val-idation, but this is very often neglected due to thedifficulty in obtaining good datasets. In a buildingenergy performance study by Diamond et al. (1992),the actual energy use was compared to the predicteduse, but this was only done for a few buildings. Shi-moda et al. (2004) presented a model that simulatescity-level energy consumption in the residential sec-tor, but these results were only compared with statis-tical data. The limited validation work on a city scaledoes not allow for the assessment of the developedmethods regarding their accuracy. Considering city-scale heating energy modeling, the sensitivity anal-ysis of the models on changes in their parameters,which is an important stage of model development,cannot really be done (Ravalico et al. 2005).

There is still a lack of study of the impact of userbehavior on urban energy performance, although asa consequence of the improved quality of thermalproperties of buildings, the role of the occupants be-comes more and more important (Santin et al. 2009).According to Shimoda et al. (2004), a city-scale ef-fect of user behavior is not really known. The workof Branco et al. (2004) also pointed out that occupantbehavior is the major factor, which causes variationin the energy consumption of different households.A “standard household,” as mentioned by Shimodaet al. (2004), is not applicable for realistic city-scalesimulation. Here again, the importance of validationof the models using real measured data is apparent,as it can provide information as to how much theoccupant behavior differs from that of the standardhousehold (Caldera et al. 2008). In most of the avail-able models, the influence of occupants on energyuse is very poorly represented (Shipworth 2010).Current approaches put standard user-behavior pa-rameters into the energy models or use so-called“occupancy schedules.”

Dow

nloa

ded

by [

Ane

ta S

trza

lka]

at 0

0:26

18

Aug

ust 2

011

528 VOLUME 17, NUMBER 4, AUGUST 2011

In order to calculate the energy demand on entiredistrict or city, it is necessary to have the requiredgeometrical information as well as the user-behaviordata of the buildings in the area of interest. On thislarge scale, it is difficult to find geometrical infor-mation with a high Level of Detail (Mavrogianniet al. 2009). Geometric data is available for mostmajor cities in the form of digital 3D city models.These models are mainly used for tourist applica-tions, urban planning, and various other fields in theurban domain. Many of the models used in thesefields are geometrical models that represent the vi-sual appearance of the real world with very little orno additional information or numeric data. Modelsfor such scenarios tend to focus on the realistic vi-sual representation that can be used to superimposeadditional information, e.g., for planning purposes.For simulation models such as the one presented inthis article, a purely geometrical model is not suffi-cient. Metral et al. (2009) also described the need forsemantically enriched 3D city models, which do notsolely focus on the geometrical/visual aspect of thereal-world environment. Especially for simulations,additional semantics and attributes are necessary. Agrowing trend from the purely visual models towarda semantically modeled urban information space canbe observed. Data formats, such as CityGML (OpenGeospatial Consortium 2008), are good examplesfor the actual modeling of urban space beyond thevisual appearance. These models include informa-tion about objects and their interrelationships as wellas attributes, which can also be linked from otherinformation systems. In this way, the 3D city modelcan act as a “base map,” which can be enhanced byall kind of data. For energy demand calculations,attributes, especially building type, heat transfer co-efficient, or year of construction, are important inputparameters for the simulation. These 3D city modelsalso need more sophisticated management systems,often called 3D-GIS (3D geoinformation systems),which can extract visual 3D models, but they alsoneed datasets with more information attached in or-der to be entered into a simulation, for example. Aprototype implementation for flexible managementof these “semantically enriched” models that wasused in the energy demand forecast scenario will bepresented in this article.

Nevertheless, the effect of urban geometry onenergy consumption still requires study. The reasonis the difficulty of modeling complex urban geom-etry (Ratti et al. 2009). For the purposes of urbanscale simulation, it is important to achieve a good

compromise between modeling accuracy, computa-tional overhead, and data availability (Robinson etal. 2009). Especially in case of 3D city models, themaximum Level of Detail is mainly restricted tothe outer building boundary and the detailed roofstructure. Modeling of facade elements and inte-riors is mostly limited to small areas in the citycenter or to specific project sides. The compromisein the described scenario is that the “low detailed”information in terms of geometry is available forthe whole city. Therefore, city-wide energy demandestimation is possible, though certain assumptionsneed to be made due to the low Level of Detail ofthe available information.

The case study district

Scharnhauser Park (SHP) is a mixedresidential–commercial area that is located onthe southern border of Stuttgart, Germany. The areais a former military area, in which office space,residential areas, and parks have been integrated.The area of SHP (Figure 1) covers 150 hectaresand currently houses 7000 inhabitants. About 80%of the heating energy demand of the whole area ofSHP is supplied by renewable energies. The mainportion of heating and electrical energy is deliveredto the buildings in SHP from a 6.3-MW thermal and1-MW electric wood-fired co-generation plant.

Methods

Data acquisition

The annual heating energy consumption data forall buildings was obtained from the archive of themunicipal utility company. Data was provided in theform of either Excel data or printouts and was thenorganized in an access database.

In three case study buildings, an automatic mon-itoring system was installed to measure the heat-ing energy consumption of the whole building andthe separate apartments within it. A GSM-modemwith an M-bus interface and integrated data loggerenables the data to be recorded at 1-h increments.These data are then automatically transferred usinga mobile phone interface, which sends the data viae-mail to the central simulation server every day.

The main thermal data needed for the simulationmodel are the heat transfer coefficients (U-values).

Dow

nloa

ded

by [

Ane

ta S

trza

lka]

at 0

0:26

18

Aug

ust 2

011

HVAC&R RESEARCH 529

Figure 1. GIS thematic map of the case study district with building footprints (color figure available online).

SHP is a modern residential area with low-energybuildings. As the buildings were all constructedwithin the last decade, in which similar legal re-quirements applied for thermal standards, the ther-mal values needed for the simulation model are com-

parable for all of the buildings. Therefore, in a firstsimplification step, these values have been assumedas averages for two building groups (row houses[RHs] and multi-family houses [MFHs]), as shownin Table 1.

Table 1. Average U-values for the buildings constructed between 2000–2008.

Building element

Building element Outer wall Roof Floor Window

Building category RHsAverage U-value, W/m2K 0.22 0.16 0.21 1.3

MFHsAverage U-value, W/m2K 0.24 0.21 0.56 1.2

Dow

nloa

ded

by [

Ane

ta S

trza

lka]

at 0

0:26

18

Aug

ust 2

011

530 VOLUME 17, NUMBER 4, AUGUST 2011

Hourly air temperature and global radiation datawas provided for 2008, 2009, and 2010 from theweather station placed on the roof of the biomasspower station in SHP.

Data framework

In the presented approach, it is necessary to ex-tract information from the 3D city model, which isnot explicitly stored in the model itself. One exam-ple is the overall area of the outer envelope of thebuilding. This value needs to be extracted from theavailable data, as well as the volume, orientation ofwalls, etc.

It is also necessary to find walls that are sharedbetween buildings and are, therefore, not exposedto wind and irradiance. Building objects in 3D citymodels are mainly modeled as single objects withno information about adjacent objects. This infor-mation needs to be extracted and provided to thesimulation tool in an appropriate way. In order toextract this additional data, a 3D data managementframework that is capable of reading 3D city mod-els and additional data from different/distributedsources was used, then the external information wasconverted into an internal data representation. Basedon the internal data format, the framework is capableof extracting the required information and linkingdata from other sources. A software tool has beendeveloped on top of the 3D framework that managesqueries of the 3D model for the area of interest, theextraction of the required information, and the cre-ation of the output data, which can be read by thesimulation tool.

GIS-interface providing input forsimulation model

The given building footprints, combined with themeasured building heights, are used to generate atopologically consistent 3D city model (see Figure2). This topologically correct model, created accord-ing to the method of Ledoux and Meijers (2009), isnot just an extrusion of each individual footprint bythe given height, but it also takes into account build-ings that share walls. The resulting dataset of thetopologically correct model included a geometricaldataset of the model encoded in CityGML (OpenGeospatial Consortium 2008) and a text file withinformation as to which wall is an outer/inner wall,as well as identifiers for roof and floor faces.

Figure 2. Topologically consistent 3D block model of SHP.

In the presented approach, it is essential for cal-culations to know which of the faces are inner/outerwalls, roofs, and floors. This kind of classificationcan only be done if the model is generated accordingto specific semantics or a specific ontology (Metralet al. 2009). Additional information, such as theyear of construction, building type, etc., can also beuseful and can be integrated into the model as partof a future study.

Based on this topologically correct 3D city model(block or detailed), the total area of outer walls,walls between buildings, ground floor, and roof(Figure 3) can be calculated for each building, asshown in Table 2 for the above example.

In order to calculate the required values, an inte-gration of several data streams was necessary. The

Figure 3. Example of the extraction.

Dow

nloa

ded

by [

Ane

ta S

trza

lka]

at 0

0:26

18

Aug

ust 2

011

HVAC&R RESEARCH 531

Table 2. Example of data calculation of the building data.

Building element

Building ID Ground floor area, m2 Internal wall area, m2 Outer wall area, m2 Roof area, m2

ID1 64 80 177 64ID2 73 138 162 73ID3 62 115 141 62ID4 53 115 96 53

framework architecture is highly modular, and ap-plications can use specific components that are suit-able for the task at hand. The framework providesmodules for three major areas: data input (connec-tors), data manipulation (data mapping), and dataoutput (creators) (see Figure 4).

The interface that generates the output for thesimulation tool is developed on top of this frame-work using several components. The frameworkis used to connect to the Environmental SystemsResearch Institute (ESRI) shape-file that includedthe original building footprints and heights of thebuildings (used for extrusion) and to connect to theCityGML file, which houses the semantic 3D citymodel. The data-mapping components are used toanalyze the 3D city model geometry, connect thewall classification information, and calculate theoverall inner/outer wall area of each building. Theinterface component writes this information into aCSV text file that can be read by the simulationmodel, which makes it possible to adjust the outputformat to possible requirements of specific simula-tion tools.

The window area in this case was obtained witha very simplified assumption as a part of the outer

Figure 4. 3D data management framework.

wall area: 12% for the RHs and 20% for MFHs. Theso-prepared geometry data for each building wasput in one data file and serves as the input for thesimulation model, according to the workflow seenin Figure 5.

Calculation and simulation model

The “GIS-interface-file,” which includes build-ing dimensions, was then entered into the buildingsimulation model. In order to achieve reliable re-sults regarding the heating energy demand with aslittle input as possible, two versions of models weretested. The first one (Model 1) calculated the heat-ing energy demand by taking into consideration onlythe transmission losses through the outer envelopeof the building. The second model (Model 2) con-sidered the entire energy balance with transmissionand ventilation losses as well as solar and internalgains following the well-established energy balancemethod described, for example, in German StandardDIN V 18599 (DIN e.V. 2010).

In both models, some assumptions were made; inModel 1, the heating set-point temperature was setto a constant value of 20◦C (68◦F), and in Model 2,the heating set-point temperature was 68◦F, the airexchange rate was 0.5 1/h, and internal heat gainswere 5 W/m2. Both models were used to calculatethe annual heating energy consumption for all resi-dential buildings in SHP. The results of heating en-ergy demand were validated by the measured annualheating energy consumption values for all buildings.

Figure 5. Workflow of data exchange.

Dow

nloa

ded

by [

Ane

ta S

trza

lka]

at 0

0:26

18

Aug

ust 2

011

532 VOLUME 17, NUMBER 4, AUGUST 2011

Table 3. Errors in demand calculation as a function of geometry and U-value precision.

Year

2009 2010

No. Calculation procedureDeviation

Model 1, %Deviation

Model 2, %Deviation

Model 1, %Deviation

Model 2, %

1 Dimensions from laser scanning andaverage U-values

−39 −22 −34 −10

2 Dimensions from laser scanning and realU-values

−29 −13 −23 −1

3 Real dimensions and average U-values −29 −19 −23 −54 Real dimensions (including real window

orientation) and real U-values−20 −10 −13 +5

Results

Comparison between measured andcalculated values

Case study MFHThe case study building is a 12-apartment MFH

with a gross heated area of 1688 m2. The annualmeasured heating energy consumption for the year2009 was 56 kWh/m2/a and for the year 2010 was57 kWh/m2/a. The deviation between these valuesand the calculated demand using the energy-balancemethod (Model 2) was 5%–10% when using the realdimensions (including the real window orientation)and thermal properties of the building taken fromthe building certificate. When the geometry deter-mination is simplified or when average U-values areused, the error increases.

Table 3 shows how the precision of the dimen-sions and the thermal properties influence the annualheating energy demand calculations. Due to the factthat all buildings are calculated using Procedure 1(dimensions from laser scanning and average U-values), a deviation of 10%–22% can be expectedfor Model 2. The calculation that uses the very sim-ple transmission losses model can even have an errorof 34%–39%.

The results depicted in Table 3 show that notonly the precision of dimensions (geometry), andmore importantly the U-values, influences the re-sults of the heating energy demand, but also themodeling technique used. The results of the steady-state energy balance of the building (Model 2) aresignificantly more accurate than those from the verysimple calculation using Model 1. Model 1, whichconsiders only the transmission losses, results in an

error of 13%–20%, even when the real dimensionsand U-values are used in the calculation.

When the time step of the calculation is reducedfrom a year to a month, good correlation betweenthe measured and calculated values can be observedin the case of the energy-balance method (Model2), see Figure 6. By summing up the monthly valuesfor the winter and summer periods, a reasonableaccuracy (error <10%) for Model 2 can be observed(see Table 4).

When a time resolution of one day is used inthe calculation, which is seen in the Figure 7,the deviation between the demand and measure-ment increases for both the degree-day method andthe energy-balance method, even when the cor-rect dimensions and U-values are used. This canbe mainly attributed to user-dependent ventilationstrategies, changing heating set-points with thermo-static valves, use of shading systems, etc.

Figure 6. Comparison between measured and calculatedmonthly values for a case study multi-family building (2009).

Dow

nloa

ded

by [

Ane

ta S

trza

lka]

at 0

0:26

18

Aug

ust 2

011

HVAC&R RESEARCH 533

Table 4. Deviation between measured and calculated values for heating periods.

2009 2010

Year Model 1 Model 2 Model 1 Model 2

Time period Heatingperiod

Summer Heatingperiod

Summer Heatingperiod

Summer Heatingperiod

Summer

Deviation in % 24 9 8 6 26 24 3 10

Figure 7. Daily measured and simulated profiles for the casestudy building.

Figure 8 shows the annual heating energyconsumption values for 10 of 12 individualapartments within the analyzed multi-family build-ing. Although the apartments have very similar en-ergy characteristics, the heating energy consump-

tion values vary quite significantly. It is quite clearthat these deviations are due to difference in userbehavior, which should be taken into account in theenergy balance Model 2.

All MFHsThe residential sector of SHP is divided into two

residential building categories: the one- and two-family RHs and the MFHs. Figure 9 shows a com-parison of the measured and calculated annual val-ues for each individual MFH.

It can be seen in Figure 9 that there is a rea-sonable correlation between the measured and cal-culated values for many of the buildings, but theanalysis also indicates some extreme deviations. Ex-treme deviations, which will be discussed in the nextsection, led to the quite high average deviations be-tween the measured and calculated values of about35%–40% by both models. Only a few buildingswith deviations greater 100% were excluded fromthis average.

Figure 8. Annual heating energy consumption of similar apartments in the multi-family case study building (color figure availableonline).

Dow

nloa

ded

by [

Ane

ta S

trza

lka]

at 0

0:26

18

Aug

ust 2

011

534 VOLUME 17, NUMBER 4, AUGUST 2011

Figure 9. Comparison of the calculated and measured annual values for all MFHs in the district (2008).

All RHsThe comparison of the measured and calculated

annual values for the heating energy consump-tion/demand of RHs, which is seen in Figure 10,also indicates many extreme deviations. Both mod-els had a quite high average deviation of over 30%.The shape of the measured values fluctuates very

strongly in comparison with the calculated values,which have a rather flat shape. The rather similarheating energy demand values for all RHs are dueto the fact that the calculation depends mainly onthe rather similar surface-to-volume ratios of thebuildings; the measured values, on the other hand,are strongly influenced by user behavior. This effect

Figure 10. Comparison of the measured and calculated annual values for all RHs (2008).

Dow

nloa

ded

by [

Ane

ta S

trza

lka]

at 0

0:26

18

Aug

ust 2

011

HVAC&R RESEARCH 535

Figure 11. Average heating consumption values and standard deviation for different types of RHs (2009).

becomes more important the smaller the buildingsare.

Therefore, actual user and occupancy behaviorof the buildings (exact time of occupancy start orreduced daily use of a building) needs to be exam-ined in greater detail. The construction types of theRHs are also quite different, meaning that the use ofone average U-value does not represent the actualsituation very well. This can be seen in the compari-son of the average heating energy consumption datafor different types of RHs in Figure 11. The averageconsumption varies from about 55 to 90 kWh/m2/a,and the standard deviation for each building type isas high as 30%.

Firth et al. (2010) mentioned in their analysisof the sensitivity of their community domestic en-ergy model, that “The heating demand temperature(which in most cases is the thermostat set-point tem-perature used in the dwelling to control the heatingsystem) results in the most sensitivity” (p. 33). Ac-cording to Shipworth (2010), internal temperatureis one of the most influential factors concerningdomestic energy use. Therefore, an attempt to ana-lyze, in detail, the influence of user-behavior-relatedparameters, like heating set-point temperature, wasperformed in this analysis. By generating randomnumbers that have a Gaussian distribution with amean value of 68◦F, a strong correlation betweenthe measured and calculated values for one type

of the RHs could be achieved. Figure 12 shows twotypes of calculations; the first was done using a fixedheating temperature set-point of 68◦F, and the sec-ond by varying the temperature set-points. The vari-ation of the temperature set-points made it possibleto achieve a good correlation between the calculatedand the measured values.

The work of Shipworth (2010) analyzed housesbuilt to identical technical specifications and ob-served that their energy consumption can vary by afactor of three once occupied. This observation wasalso made in the research work presented here (seeFigure 12).

All residential sectors of SHPFigure 13 shows the annual sum of the calculated

heating energy for all RHs and MFHs, which isin quite good agreement with the measured data,especially with that from the RHs. Here the user-dependent factors relevant for an individual buildingcalculation average out.

By summing up the values of all of the build-ings in SHP for each hour of the year, a total de-mand profile of this area could be calculated, di-vided into demand only for heating; demand forheating and warm water; and demand for heating,warm water, and heat losses from the district heat-ing system (the average annual value in 2008 forheat losses is 17%). The comparison of the last

Dow

nloa

ded

by [

Ane

ta S

trza

lka]

at 0

0:26

18

Aug

ust 2

011

536 VOLUME 17, NUMBER 4, AUGUST 2011

Figure 12. Comparison of the measured and calculated values for one type of RH. The points use a fixed heating temperatureset-point of 20◦C (68◦F); the dotted curve has variable temperature set-points.

demand profile and the measured profile of the heat-ing energy supply from the biomass co-generationsplant in SHP shows a reasonable correlation, whichcan be seen in Figure 14. The total difference be-tween the measured annual energy supply and thecalculated demand without heat losses is about20%.

Discussion

Explanation of the uncertainties

The reason for the discrepancies between mea-sured and calculated values can be seen in the Levelof Detail of the 3D city model that was used. Eight

Figure 13. Overall annual comparison of measurement and demand for two building groups: MFHs and RHs.

Dow

nloa

ded

by [

Ane

ta S

trza

lka]

at 0

0:26

18

Aug

ust 2

011

HVAC&R RESEARCH 537

Figure 14. Annual measured and calculated load duration curves for the whole district (2008).

percent of the buildings in SHP have a flat roof;therefore, in this case, the block model (Level of De-tail 1 [LOD1]) seemed sufficient for the extractionof the building heating volume. For the remaining20% of buildings with sloped roofs, LOD1 is notprecise enough for this purpose. For the buildingsbuilt after 2002, no laser scanner data are available,and therefore, the height value was set to 7.5 m.This could lead to building heating volumes that aretoo low. Furthermore, in many cases, the heatingenergy consumption value is not measured for eachbuilding individually but for a group of buildings(with shared energy meters) and then divided by thebuilding group area. This fact results in inaccuratevalues for individual buildings. The approximationof the heating consumption for warm water, whichwas set at a constant value of 12.5 kWh/m2/a andsubtracted from the total annual heating energy con-sumption values, also causes deviations between themeasured and calculated values.

Comparison with other available methods

The heating energy demand of residential areascan be estimated using several methods. The methodpresented here considers not only the city scale, asis done in district typification methods, but also the

building scale. Here, the heating energy demandis calculated for each building separately, and thenthe heating energy demand of the entire district isobtained by adding up all of these individual values.

As the buildings in the SHP study area all havea very high energy standard, it was expected thattheir annual heating energy consumption would notbe greater than 70 kWh/m2/a. In reality, the analysisshowed that although the heating energy consump-tion of most of the residential buildings was withinthe range of 40–80 kWh/m2/a, there are also manybuildings with much greater energy consumption.

Although the method presented in this article isvery similar to the method of Blesl (2010), the re-sults of the heating energy demand values could beconfirmed with monitored heating energy consump-tion values. The validation process of the calcu-lated values made it possible to identify deviationsbetween the measured and calculated values. Theanalysis of one case study MFH has proven that theLevel of Detail of the input data regarding dimen-sions and U-values influences the results greatly.The further deviations are due to different user be-havior; analysis of similar buildings and apartmentsshowed significant differences between the heatingenergy consumption of the analyzed objects. Theworks of Robinson et al. (2009) and Page (2007)

Dow

nloa

ded

by [

Ane

ta S

trza

lka]

at 0

0:26

18

Aug

ust 2

011

538 VOLUME 17, NUMBER 4, AUGUST 2011

indicated that the occupant’s behavior is one of thekey sources of the accuracy of building and urbansimulation.

These deviations will be further investigated inthe future. Due to the fact that the modeling tech-nique plays an important role, when the correct di-mensions and thermal data are known, a fully dy-namic simulation will be done. The results of thismodel will be compared to the results of the energy-balance model, in order to show whether or notModel 2 (less input data) will be sufficiently ac-curate for predicting the heating energy demand ifuser behavior is taken into account.

Summary

This article presents a method of forecasting theheating energy demand of a whole city quarter andvalidates the models with measured data. Two mod-els were tested and validated by comparing theirresults with the actual heating energy consumptionof the buildings as given in actual utility bills. Oneprimary goal was to test how effectively a 3D modelcould estimate the heating energy use. The presentedapproach shows that 3D city models are useful forurban scale simulations. However the models needto be generated in a semantically enriched way andaccording to a specific ontology (e.g., CityGML).Purely geometric models do not suffice to satisfythe information demand of simulation tools. In theauthors’ view, the geometrical models are widelyavailable as many municipalities have already pro-duced them. These were indeed not created withthe main intention of using them for simulation, butthey might be used as a basis for a semantically en-riched model. Some municipalities already have se-mantically built models based on CityGML, and theintegration of additional information in these casesis much easier. The presented approach outlines thefeasibility of connecting 3D city models/3D GISwith simulation tools and shows that further re-search into this field would be beneficial.

Depending on the assumptions that are made asto the input data (U-values and geometrical informa-tion), there are varying levels of agreement betweenthe measured and calculated annual heating values.When the calculation was performed using real U-values and dimensions, the deviation between themeasured and calculated values was about 10% withan energy-balance model (Model 2) and about 20%with a transmission-loss model (Model 1). Whenaverage U-values and general dimensions (ground

floor multiplied by building height) were used forthe calculation, the deviation increased to 22% withModel 2 and to 39% with Model 1. As most ofthe buildings were calculated by simplifying the di-mensions and using average U-values, the averagedeviation between the measured and calculated val-ues was about 35%–40%.

The analysis showed that both of the tested mod-els were suitable for forecasting the city-scale heat-ing energy demand, but the energy balance modelwas better because it also considers the ventilationlosses and solar and internal gains. Regarding thedata inputs, both models require the same Levelof Detail regarding the building dimensions andthermal parameters. The only additional informa-tion that is needed by Model 2 is the orientationof the windows, as this model also takes solar gainsinto consideration. Another advantage of Model 2 isthe possibility of varying the user-behavior-relatedparameters, such as the set-point temperature, in-ternal heat gains, air exchange rate, and duration ofheating. It could be shown that statistical variationsof heating set-point temperatures strongly improvethe accuracy of the building modeling, the more sothe smaller the building is.

Acknowledgments

The authors would like to thank the LANDESS-TIFTUNG Baden-Wurttemberg (Project: EnergyEfficient Cities) and the EU-Project POLYCITY(REF EC: TREN/05FP6EN/S07.43964/513481/)for funding part of this research. The authors wouldlike to thank Hugo Ledoux and Martijn Maijersfrom TU Delft for their help generating the topo-logically correct 3D city model. Special thanks alsoto Martin Huber (from zafh.net) for his help with theautomated transfer of the heating energy consump-tion data via the M-bus data logger and also to DirkPietruschka and Jurgen Schumacher (from zafh.net)for the technical support with the INSEL-simulationmodel. The authors would like to acknowledge fi-nancial support of the CITYNET project funded viathe Marie Curie Research Training Network. Thisproject financed the presentation of this article atthe IAQVEC 2010 conference in Syracuse.

ReferencesAutodesk White Paper. 2011. Streamlining energy analysis of

existing buildings with rapid energy modeling. San Rafael,CA.

Dow

nloa

ded

by [

Ane

ta S

trza

lka]

at 0

0:26

18

Aug

ust 2

011

HVAC&R RESEARCH 539

Blesl, M. 2002. Raumlich hoch aufgeloste modellierungleitungsgebundener energieversorgungssysteme zur deckungdes niedertemperaturwarmebedarfs (Spatial high-resolutionmodeling of the pipeline-bound energy supply systems forcovering low temperature heating energy demand). PhD dis-sertation, Stuttgart University, Stuttgart, Germany.

Blesl, M., S. Kempe, and H. Huther, 2010. Capture of spatialhigh-resolution building heating energy demand. Euroheat& Power 39(1-2):28–33. [in German]

Branco, G., B. Lachal, P. Gallinelli, and W. Weber. 2004. Pre-dicted versus observed heat consumption of a low energymultifamily complex in Switzerland based in long-term ex-perimentail data. Energy and Building 36:543–55.

Caldera, M., S.P. Corgnati, and M. Filippi. 2008. Energy demandfor space heating through a statistical approach: Applicationto residential buildings. Energy and Buildings 40:1972–83.

Diamond, R.C., M.A. Piette, B. Nordman, O.deBuen, and J. Har-ris. 1992. The performance of the energy edge buildings:Energy use and savings. Proceedings of ACEEE 1992 Sum-mer Study on Energy Efficiency in Buildings, Berkeley, CA,3:3.47–60.

DIN e.V. [German Institute for Standardization]. 2003. DIN EN12831—Heating systems in buildings—Method for calcula-tion of the design heat load, 12831:2003. [in German]

DIN e.V. 2010. DIN V 18599—Energy efficiency ofbuildings—Calculations of the net, final and primary energydemand for heating, cooling, ventilation, domestic hot wa-ter and lighting—Supplement 1: Balancing of demand andconsumption. [in German]

Firth, S.K., and K.J. Lomas. 2009. Investigating CO2-emissionreduction in existing urban housing. Proceedings of BuildingSimulation 2009, 11th IBPSA Conference, Glasgow, Scotland.

Firth, S.K., K.J. Lomas, and A.J. Wright. 2010. Targeting house-hold energy-efficiency measures using sensitivity analysis.Building Research & Information 38:25–41.

Jaffal, I., Ch. Inard, and Ch. Ghiaus. 2009. Fast method to predictbuilding heating demand based on the design of experiments.Energy and Buildings 41:669–77.

Ledoux, H., and M. Meijers. 2009. Urban and Regional DataManagement, UDMS Annual 2009, Extruding building foot-prints to create topologically consistent 3D city models. A.Krek, M. Rumor, S. Zlatanova, and E. Fendel, eds, 39–48.Ljubijana, Slovenia: CRC Press.

Mathews, E.H., Y. Etzion, E. Heerden, S. van Weggelaar, E.Erell, D. Pearlmutter, and I.A. Meir. 1997. A novel thermal

simulation model and its application on naturally ventilateddesert buildings. Building and Environment 32(5):447–456.

Mavrogianni, A., M. Davies, M. Kolokotroni, and I. Hamilton.2009. A GIS-based bottom-up space heating demand modelof the London domestic stock. Proceedings of 11th Interna-tional IBPSA Conference, Building Simulation 2009, Glas-gow, Scotland.

Metral, C., G. Falquet, and A.F. Cutting-Decelle. 2009. Towardssemantically enriched 3D city models: An ontology-based ap-proach. Academic Track of GeoWeb 2009—Cityscapes, Inter-national Archives of Photogrammetry, Remote Sensing andSpatial Information Sciences (ISPRS), XXXVIII-3-4/C3, Van-couver, Canada.

Open Geospatial Consortium. 2008. OpenGIS City GeographyMarkup Language (CityGML) Encoding Standard, Wayland,MA. G. Groger, T. Kolbe, A. Czerwinski, and C. Nagel, eds.

Page, J. 2007. Simulating occupant presence and behaviour inbuildings. Ph.D. thesis no. 2900, Politechnic of Lausainne,Switzerland.

Ratti, C., N. Baker, and K. Steemers. 2005. Energy consumptionand urban texture. Energy and Buildings 37:762–76.

Ravalico, J.K., H.R. Maier, G.C. Dandy, J.P. Norton, and B.F.W.Croke. 2005. A comparison of sensitivity analysis techniquesfor complex models for environmental management. Pro-ceedings of MODSIM05, Melbourne, Australia, pp. 2533–9.

Robinson, D., F. Haldi, J. Kampf, P. Leroux, D. Perez, A.Rasheed, and U. Wilke. 2009. CITYSIM: Comprehensivemicro-simulation of resource flows for sustainable urbanplanning. Proceedings of 11th International IBPSA Confer-ence, Building Simulation 2009, Glasgow, Scotland.

Santin, O.G., L. Itard, and H. Visscher. 2009. The effect of oc-cupancy and building characteristics on energy use for spaceand water heating in Dutch residential stock. Energy andBuildings 41:1223–32.

Shimoda, Y., T. Fujii, T. Morikawa, and M. Mizuno. 2004. Resi-dential end-use energy simulation at city scale. Building andEnvironment 39:959–67.

Shipworth, D. 2010. The vernacular architecture of householdenergy models. Modelling in the Social Sciences: Interdisci-plinary Comprison, Helsinki Collegium for Advances Stud-ies, June 17–18, 2010.

Swan, L.G., and V.I. Ugursal. 2009. Modeling of end-use energyconsumption in the residential sector: A review of model-ing techniques. Renewable and Sustainable Energy Reviews13:1819–35.

Dow

nloa

ded

by [

Ane

ta S

trza

lka]

at 0

0:26

18

Aug

ust 2

011