Comparison of prediction models for determining energy demand in the residential sector of a country

Energy and Buildings 128 (2016) 38–55

Contents lists available at ScienceDirect

Energy and Buildings journal homepage: www.elsevier.com/locate/enbuild

Comparison of prediction models for determining energy demand in the residential sector of a country Aner Martinez Soto ∗ , Mark F. Jentsch Urban Energy Systems, Faculty of Civil Engineering, Bauhaus-Universität Weimar, Weimar, Germany

a r t i c l e

i n f o

Article history: Received 21 December 2015 Received in revised form 30 May 2016 Accepted 21 June 2016 Available online 21 June 2016 Keywords: Residential energy demand Energy models Model comparison Sensitivity analysis

a b s t r a c t The increasing need for energy conservation has led to the development of a range of energy models for assessing energy demand in the residential sector of a country. Even though such models deliver a principal solution for forecasting energy demand and assessing the impact of future energy saving measures, collecting the required baseline data is fraught with difficulties such as a complete lack of data, missing data within a dataset and a lack in coherence between different datasets in terms of detail, data collection method, baseline assumptions and sample size. This paper analyses the transferability and accuracy of twelve energy models (MAED-2, FfE-Gebäudemodell, CDEM, REM, CREEM, ECCABS, REEPS, BREHOMES, LEAP, DECM, CHM, BSM), taking Germany as case study example. Furthermore, a sensitivity analysis is conducted for each model to analyze the significance of the input variables for the overall modelling outcome, highlighting the most influential variables. It is shown that models with a high level of disaggregation do not necessarily guarantee more accurate results. Adjustments are proposed to improve the transferability of the models to the case study country Germany. © 2016 Elsevier B.V. All rights reserved.

1. Introduction Over the last two decades, the world’s total primary energy use across all sectors (industrial, residential, commercial, transport, agricultural) has grown by 49% with an average annual increase of about 2% [1] and currently stands at 560 EJ per year [2]. A considerable part of the global energy use is attributable to the residential sector. In the European Union (EU), the residential sector is responsible for about 26% of the total final energy use [3] and it is estimated that, overall, the residential sector represents about one third of the world’s energy use [4]. However, on a national level the share varies depending on the conditions of the individual country and can reach up to 50% for selected countries [5,6]. The progressive increase and the overall high level of energy use in the residential sector contribute to the depletion of fossil fuel resources, put stress on the supply side of energy services, potentially create energy security issues and have environmental impacts locally as well as globally [7]. Therefore, it is generally agreed that energy saving policy measures in this sector present a great potential for reducing energy use and hence CO2 emissions related to the

∗ Corresponding author at: Urban Energy Systems, Faculty of Civil Engineering, Bauhaus-Universität Weimar, Coudraystrasse 7, 99423 Weimar, Germany. E-mail addresses: [email protected], [email protected] (A. Martinez Soto). http://dx.doi.org/10.1016/j.enbuild.2016.06.063 0378-7788/© 2016 Elsevier B.V. All rights reserved.

combustion of fossil fuels [8]. The residential sector can, therefore, provide an important contribution to meeting climate and energy conservation targets such as the “20-20-20” target in the EU, which, by 2020, aims at a 20% reduction in greenhouse gas (GHG) emissions from a 1990 baseline, a 20% share of renewable energy in the final energy use and a 20% energy efficiency improvement compared to projections made in 2007 for 2020 [9–11]. However, for the evaluation of energy saving policies as well as incentives for meeting greenhouse gas emission targets, information about the potential future development of the energy demand as well as the variables that affect the actual use and could potentially influence future demand is needed [12,13]. In recent years various models have been developed for this very reason, which allow for a quantitative assessment of both, the current energy use as well as, based on the impact of different policy measures, a prediction of the future energy demand of a country [14]. Energy models where the calculation routines are publicly available and that are applicable to the residential sector include the following twelve models: MAED-2 [15], FfE-Gebäudemodell [16], CDEM [17], REM [14], CREEM [18], ECCABS [19,20], REEPS [21], BREHOMES [22], LEAP [23], BSM [24], DECM [25] and CHM [26]. These models, which are detailed in Table 1, are assessed in the following, looking at the underlying modelling approach as well as their performance for

A. Martinez Soto, M.F. Jentsch / Energy and Buildings 128 (2016) 38–55

Fig. 1. Share of the residential sector in the national final energy use according to final use. (Data sources: [35–46], Note: The baseline years for the data are 2013 for the UK, Germany, the USA and Australia; 2011 for Italy and Spain; 2010 for Canada; 2009 for South Africa and France and 2008 for Chile and China. In the case of Canada ‘Cooking’ is included under “Lighting and Appliances”. For France, Germany, Italy and the UK “Space cooling” is included under “Lighting and Appliances”).

predicting the energy demand in the residential sector for the case of Germany.1 2. Issues with current energy models for predicting energy demand in the residential sector Although residential energy models are an appropriate solution for predicting energy demand and assessing the impact of future energy saving measures in this sector, the diversity of existing modelling methods makes it difficult to select an appropriate model for assessing the development in a given country. Currently, there are a limited number of energy models which can, in principle, be used for any country. This includes, for example, the models LEAP [23] and MAED-2 [15]. However, considering that each country represents a particular case in terms of the share of the residential energy use compared to the total energy use as well as the share of different final energy uses within the residential sector, the results produced with these models may not be equally valid for all countries. This problem is highlighted in Fig. 1 which shows that there are large variations in the share of final energy uses in the residential sector, depending on the climate, economy, living standard, lifestyle and equipment that is used in different countries. Therefore, for a number of countries individual models have been developed that are specifically tailored towards the local conditions with respect to their computational routines. The models BREHOMES [22] for the UK and CREEM [18] for Canada are examples for this approach. However, it remains unclear, whether such models that were developed for a specific country are transferrable and what adjustments would be necessary to obtain an accurate estimate of the energy demand in a different country than the one that they were designed for. In addition, each model has a different level of disaggregation and, therefore, a different level of detail is needed in the input data. Models with a high level of disaggregation allow a greater understanding of the influence of various input parameters. However, the data collection which is usually done through statistical analyses [5] may, in some cases, be fraught with difficulties due to the difference

1

Note: For evaluating the model outputs presented here the term “energy use” denotes the statistically determined use of energy in the past, whereas “energy demand” specifies the future energy needs on the basis of the modelling. This distinction follows the recommendations by Chateau and Lapillonne [27].

39

in detail of the underlying data bases [16] or, in other cases, may force policy developers to put a further level of detail to the existing baseline information [37]. This ultimately slows the modelling process. By contrast, models that do not have a high level of detail may underestimate or even completely miss the impacts of a specific energy saving measure. This could potentially have an adverse effect on decision making when assessing the predicted implications of a policy measure. For example, prediction models like LEAP [23], or the FfE-Gebäudemodell [16] cannot take into account a reduction in energy demand that is induced by an improved thermal insulation. However, in the case of Germany, the continued development in the field of thermal insulation materials [47] and the introduction of the so-called “Wärmeschutzverordnung” (Thermal Insulation Regulation) in 1977 with amended versions in 1984 and 1995 and the subsequent “Energieeinsparverordnung” (EnEV, Energy Saving Regulation) published in 2002 with a major revision in 2007 resulted in a continuous decline in the average final energy use in new buildings over the last decades [48,49].2 This reduction in final energy use is mainly related to reduced space heating requirements [50] as a result of both technology changes in space heating provision and improved thermal insulation standards. This highlights the need for defining the optimum input variables to facilitate simulation and at the same time maintain a high level of accuracy in the predictions. The accuracy of the currently available residential energy models has typically been tested, but most of the existing models, apart from the models CDEM [17], DECM [25], CHM [51] and BSM [24], have not been subjected to studies looking at the effect of the input variables on the uncertainty in model prediction [25,51]. The accuracy of the models was, in general, determined for specific countries with this assessment being mostly limited to the country for which they were initially developed, comparing the prediction results with national statistical data [17]. However, this approach does not give any indication of the ability of these models to predict energy demand in other countries which raises the question of model transferability. The model uncertainty for the CDEM, DECM, CHM and BSM models was quantified through a sensitivity analysis which identified the input variables with the greatest effect on the models’ outputs [17]. Firth et al. [17], Cheng and Steemers [25], as well as Kavgic et al. [52] all agree that the current shortcomings, which lie in the often missing quantification of inherent uncertainties and the lack in transparency of the models, must be resolved and that without rigorous testing the predictions of energy models run risk of lacking credibility. The aim of this paper is, therefore, to provide a comparison of the accuracy and transferability of the twelve existing prediction models highlighted in Table 1 for determining a country’s current and future energy demand in the residential sector. This is undertaken through objective analysis parameters, such as the relative deviation error, average percentage difference, Pearson’s correlation coefficient r and the coefficient of determination r2 , taking Germany as a case study example. This includes a discussion of the present strengths and weaknesses of each model. The model comparison through objective analysis parameters furthermore gives an initial indication of the transferability of the models, most of which were not developed for the residential sector in Germany. In addition, through a sensitivity analysis of each model, the relevance of the input parameters of these twelve energy models is being assessed.

2 Note: Up to reunification in 1990 these regulations were only valid in the western part of the country. In East Germany the corresponding regulations were TGL 28706 “Bautechnischer Wärmeschutz” (Structural Heat Insulation) that was legally binding from 1976 to 1981 and TGL 35424 “Bautechnischer Wärmeschutz” (Structural Heat Insulation) from 1981 onwards.

40

A. Martinez Soto, M.F. Jentsch / Energy and Buildings 128 (2016) 38–55

Table 1 Energy models for estimating energy demand in the residential sector of a country. Model Name

Developer

Model Description

LEAP (Long-range Energy Alternatives Planning)

SEI—Stockholm Environment Institute, Tellus Institute, USA [23]

FfE Gebäudemodell (Forschungsstelle für Energiewirtschaft- Gebäudemodel)

Forschungsstelle für Energiewirtschaft e.V. (FfE), Germany [16]

REM (Regional Engineering Model)

University of Joensuu, Finland [14]

BSM (bottom-up Building-Stock-Model)

Karlsruhe Institute of Technology, Germany [24]

MAED-2 (Model for Analysis of Energy Demand)

International Atomic Energy Agency, Austria [15]

Integrated modelling tool that can be used for the analysis and evaluation of energy policies and the assessment of climate change mitigation measures across all sectors of an economy (housing, commercial, transportation and industry sectors). On the basis of available statistical data and distribution keys (floor space per building type, location of buildings, specific energy demand for space heating and hot water provision) this model allows to derive detailed conclusions for the energy demand in the German residential sector. Regional building stock model for assessing the energy demand for space heating and the related GHG emissions and costs. The model was validated using the province of North Karelia, Finland as case study area. Modular tool for predicting the building stock development (based on deterministic projections of floor space and demolition/refurbishment rates) and the energy demand (based on statistical data for the specific energy demand for space heating). The model is designed for analyzing the German federal government’s energy policy targets for the residential sector. Based on medium- to long-term scenarios of socio-economic, technological and demographic developments this model allows the prediction of the future energy demand. Calculations are performed for the housing, commercial, transportation and industry sectors.

ECCABS (Energy, Carbon and Costs Assessment of Building Stocks)

Chalmers University of Technology, Sweden [19,20,28]

CDEM (Community Domestic Energy Model)

Department of Civil and Building Engineering, Loughborough University, UK [17]

CREEM (Canadian Residential Energy End-use Model)

Canadian Residential Energy End-use Data and Analysis Centre, Canada [18]

BREHOMES (Building Research Establishment Housing Model for Energy Studies)

Building Research Establishment (BRE), UK [22]

REEPS (Residential End-use Energy Planning System)

Electric Power Research Institute, USA [21,31,32]

DECM (Domestic Energy and Carbon Model)

The Martin Centre for Architectural and Urban Studies, Department of Architecture, University of Cambridge, UK [25]

CHM (Cambridge Housing Model)

Cambridge Architectural Research Ltd (CAR), UK [26]

Calculation of the annual energy demand, CO2 emissions and energy costs associated with the residential sector. This Model is designed to assess the effects of energy saving measures and CO2 emission mitigation strategies. Using the core calculation engine BREDEM-8 (Building Research Establishment Domestic Energy Model) this model calculates the monthly energy demand and CO2 emissions in the English residential sector, disaggregating the English housing stock into 47 house archetypes. Calculation of the annual and monthly energy demand and CO2 emissions in the Canadian residential sector. The Software HOT2000 [29] is used as calculation engine. Disaggregation of the UK housing stock into over 1000 dwelling categories and calculation of the energy demand for each dwelling category with BREDEM-12 [30]. Evaluation of future trends in energy demand (based on forecasts for appliance installations, operating efficiencies, and utilization patterns for space heating, water heating, air conditioning and cooking) in the US housing sector, taking into account various user-defined assumptions and/or different development scenarios. Prediction of the energy demand and CO2 emissions of the existing English housing stock, using the core calculation engine SAP-2005 (Standard Assessment Procedure) [33]. An occupancy pattern is added to this model as a as novel feature. Based on SAP-2009 [34] the model generates estimates for the household energy demand for the UK Government’s Department of Energy and Climate Change (DECC).

A. Martinez Soto, M.F. Jentsch / Energy and Buildings 128 (2016) 38–55

3. Classification approaches for residential energy models Generally, energy models can be classified into top-down and bottom-up models according to the general approach that is applied for setting them up. In the top-down modelling approach, the system to be modeled is initially regarded as an entity, without any consideration of the underlying factors that may have an influence on the key parameters of this system [8]. In a next step each part of the system design (e.g. price effects, income effects, or investment effects) is then inferred from the key input parameters. The resulting parameters are then refined into the next level of factors that further detail the system. The system is refined until a clear overview is possible. In the bottom-up modelling approach the opposite methodology is pursued. In a first step, all the individual factors that may have an influence on the overall system are recorded in detail [53]. These are then joined together to form larger components by grouping parameters that are dependent on each other. The components are then linked to each other to create the entire model of the system. The advantage of the top-down modelling approach is that there is only a need for a limited amount of statistical data that needs to be aggregated to form the model [54]. Furthermore, the required data is generally readily available [5]. However, the lack of detail does not permit an identification of the influence that more detailed data input parameters may have on the overall model and hinders a complete understanding of the problem being analyzed [5,8,54]. The strength of the bottom-up modelling approach is that is possible to determine the influence of detailed input variables on the model output [8,54,55]. However, it has the disadvantage that the required detailed input data may not be readily available [5]. 3.1. Statistical and building physics bottom-up energy models The models analyzed in this study all use the bottom-up approach. However, they have different calculation methods, data input requirements and levels of disaggregation. Bottom-up models can be divided into two sub-groups (statistical or building physics models) depending on whether statistical information is being used for the input variables, or whether calculations are being performed within the model based on physical input data [8]. The “statistical” bottom-up approach which is used in five of the twelve investigated models (Table 2) for example considers representative dwellings that are associated with historic statistical data for energy use, such as the kWh/m2 a data published by the Deutsche Energie-Agentur (DENA) for various typical dwelling types in Germany [56]. Such historical data are then used in combination with regression analyses that are conducted on the basis of this information to predict the energy demand in the future. Examples for this type of model are: LEAP, FfE-Gebäudemodell and REM. In the “building physics” bottom-up approach the energy demand of for example space heating for a dwelling type is directly calculated according to a balance between energy gains (solar gains, occupants, equipment and appliances) and thermal energy losses (heat transfer, climate conditions, infiltration and ventilation) with a core calculation engine such as for example BREDEM-12 for BREHOMES, EABS for ECCABS and HOT2000 for CREEM. In such a core calculation engine the energy demand for hot water provision is then calculated in relation to the hot water demand which is either defined by the number of occupants or the specific power demand for hot water per dwelling area and the heat loss in the water systems from the hot water cylinder and the distribution pipework. The energy demand for cooking, lighting and appliances is principally determined by the number of occupants and the occupancy profile, whereas the energy demand for space cooling is typically calculated in relation to the time when the indoor air tempera-

41

ture exceeds a predefined upper comfort limit. The core calculation engines are either designed to calculate the hourly (DOE-2, EABS), monthly (HOT2000) or annual energy demand (BREDEM-12). All models which use this approach, i.e. bottom-up building physics models, have a limited set of model dwellings (dwellings types) that represent classes of dwellings in the overall residential sector.

3.2. Comparison of major equations in the investigated residential energy models Whilst there are similarities in the way that the majority of models manage their input parameters, they use different baseline databases and have different calculation routines for determining the energy demand. In this regard, a detailed comparison of all calculation routines is complex because there is no common approach with respect to the structure or sequence of equations. For example, to determine the number of dwellings, the “statistical” models MAED-2 and BSM use completely different input parameters. Whilst MAED-2 relates the development in the number of dwellings to the growth rate of the population (Eq. (1), adapted from [15]), BSM determines the number of buildings based on the growth rate of new builds reduced by the number of demolished buildings (Eq. (2), following [24]).



DW tot =

P(0) · 1 +

 GR n 100

(1)

Pdw

where: DWtot : Number of dwellings (can be broken down into urban and rural) P (O): Population in the starting year GR: Average annual growth rate of the population between the starting and current model years n: Number of years between the current and the starting model year P dw : Average household size (persons per dwelling) Stot s,r (t + 1) =



(Sa,s,r (t) − dospa,s,r (t) − drepa,s,r (t))

a

+

t 

adds,r (n) +

n=2010

t 

reps,r (n)

(2)

n=2010

where: Stot s,r (t + 1): Total building stock for building size category s and region r in year t + 1 Sa,s,r (t): Stock for building age category a, building size category s and region r in year t dospa,s,r (t): Demolition for open space use for building age category a, building size category s and region r in year t drepa,s,r (t): Replaced demolition for building age category a, building size category s and region r in year t add s,r (t): Additional new build for building size category s and region r in year t reps,r (t): Replaced new build for building size category s and region r in year t Similar to MAED-2 the “building physics” model BREHOMES establishes the number of dwellings on the basis of projections for the population development and the average household size [22]. However, in this case, in a first step BREDEM, i.e. the core calculation engine of BREHOMES, is used to establish the average number of occupants per dwelling. For this two equations are applied in which a relationship between the number of occupants and the floor area

42 Table 2 Comparative summary of the main characteristics of the twelve energy models considered in this study, sorted by their modelling approach into bottom-up statistical and bottom-up building physics models (see also Table 1). Bottom-up building physics models

Bottom-up statistical models

Application Universal Specific country (nation)

X X

Sector of application Residential Other sectors

X X

Level of original disaggregation Not determineda

a b c

REM

ECCABS

CDEM/BREHOMES

DECM/CHM

CREEM

REEPS

X (Germany)

X (Finland)

X (Sweden)

X (England/UK)

X (England/UK)

X (Canada)

X (USA)

X

X X

X

X

X X

X X

X X

X

X

X

X

X

X

X

80 house archetypesb /70 house archetypesb

4163 unitsc

300 house archetypesb

47 house archetypesb /1000 house archetypesb

50 house archetypesb /14951 representative dwellings

8767 house archetypesb

8 house archetypesb

X

X X

X X

X X

X X

X X

X

X

Spatial coverage Regional National

Data output Final Energy CO2 emissions

BSM/FfEGebäudemodell

X only Leap

Due the universal nature of these models it is not possible to determine a specific level of disaggregation, as the modeler has to specify the disaggregation himself. The term “house archetype” used in this paper encompasses various terms originally used by the developers of the models: building categories, house archetypes, housing types, and building types. “Units” are municipally aggregated groups of buildings with similar space heating demand features.

A. Martinez Soto, M.F. Jentsch / Energy and Buildings 128 (2016) 38–55

LEAP/MAED-2

A. Martinez Soto, M.F. Jentsch / Energy and Buildings 128 (2016) 38–55

of the dwelling is being established (Eqs. (3) and (4) following [30]). From this Eq. (5) for the total number of dwellings can inferred. N = 0.0365TFA − 0.00004145 · TFA2 for TFA >

450m2

(3) (4)

P (t + 1) · DWtot (t)

k

i=1

(5)

Ni · DWi

where: TFA: Total floor area of dwelling N: Number of occupants per dwelling (can be broken down into dwelling type i, Ni in starting year t) P (t + 1): Projections for the population in year t + 1, P(t) = Population in starting year t DWtot (t): Total number of dwellings in the starting year t DWi : Number of dwellings per dwelling type i in starting year t Similar disparities are observed in the equations for determining the energy used for different end-uses (heating, hot water provision, cooking, lighting and appliances). For example, for determining the energy demand for heating and hot water provision the “statistical” models FfE-Gebäudemodell and BSM employ the average specific energy demand per floor space area [16,24]. However, due to the use of different baseline databases, a direct comparison of results is difficult, as building types of different age groups with different reference values, i.e. heating energy demand (FfEGebäudemodell) and final energy demand (BSM), are used. This results in different model outputs despite the fact that they essentially share the same calculation routines. “Building physics” models use individual core calculation engines so that a detailed comparison of the calculation routines is more complex than with “statistical” models. For example, BREDEM-12, i.e. the core calculation engine of BREHOMES, estimates the annual energy demand for hot water provision considering the number of occupants and the energy losses through the primary pipework, the hot water cylinder and the distribution pipework (Eq. (6), following [30]). By contrast, EABS, i.e. the core calculation engine of ECCABS, determines the hot water demand considering the average specific power demand for hot water per floor space area (W/m2 ) (Eq. (7), adapted from [19]). Qhotwd = {Q u + Qpp + Qt + Qd − Qs }/(31, 71 · εw )

[GJ/a]

(6)

where: Qhotwd : Total energy demand for hot water production Qu : Hot water demand related to the number of occupants Qpp : Primary pipework losses Qt : Hot water cylinder losses Qd : Distribution pipework losses Qs : Solar panel contribution to hot water production εw : Efficiency of hot water heating system Qhotwd =

1 3.6 · 10

6

·

for being integrated into an improved residential energy model in the future. 3.3. Approaches for a more detailed classification

9 N=   1 + 54, 3 1+ TFA DWtot (t + 1) =

for TFA ≤ 450m2

43

8760 0

Hw · A · 3600

[kWh/a]

In order to facilitate comparison, various attempts have been made for categorizing energy models. These categorizations are, however, using different criteria and, according to the criteria adopted, the models may belong to different categories in different studies. For example, Nakata analyzed energy economic models and determined LEAP to be a “Fixed Coefficient Model” [57], whereas Bhattacharyya and Govinda classified the same model (LEAP) as a “Bottom-up Optimisation/Accounting Model” [58]. In this study, the criteria used to classify the given twelve energy models are mainly based on the work of Kavgic et al. [8] and Swan and Ugursal [5]. This is due to the amount of up to date information which these studies cover and the fact that the four energy models listed in these works (BREHOMES, CREEM, REM, CDEM) are also analyzed in this paper. Table 2 gives an overview of the main characteristics of the twelve energy models assessed in this work, grouping together the models with similar characteristics. As already indicated in Section 2, energy models can be classified according to their application into universally applicable models that have the ability to determine the current energy use/future energy demand of any country and country specific models that were created to estimate the energy use/demand of a single country. As shown in Table 2 two of the 12 analyzed models can be universally applied (MAED-2 and LEAP), whilst the remaining ones were created to determine the energy use/demand in a particular country (BREHOMES, CDEM, CREEM, ECCABS, FfE-Gebäudemodell, BSM, REEPS, CHM, DECM and REM). It has been shown that variations in energy use at a regional level (e.g. related to energy demand per m2 according to dwelling type) can strongly influence the national average energy use in the residential sector [59]. For this reason there is currently a trend in model development towards breaking down the residential energy use into regions, zones or even cities [25,60]. This is also shown in Table 2 which highlights that seven of the more recent models (LEAP, MAED-2, REM, DECM, CHM, CREEM and REEPS) are suitable for calculating the energy use at a sub-national level. The advantage of this approach is that it allows for the development of local energy policies related to a specific city or region. In addition, this type of model allows for a subdivision into the regional shares of the national energy use according to the various climate regions of a particular country. The amount of data required due to the breakdown of the input data to the local level may, however, hinder the modelling if an entire country is to be assessed. 3.4. The role of the level of disaggregation for the modelling

(7)

Hw : Average specific power demand for hot water production (W/m2 ) A: Area of heated floor space (m2 ) The above highlights the complexity of conducting an exhaustive comparison regarding the calculation routines of the various residential energy models investigated in this work which would go beyond the scope of this paper. However, it also demonstrates the importance of assessing model accuracy and transferability as well as conducting sensitivity analyses on them in order to determine which input parameters and equations have the highest potential

It should be noted that most of the models analyzed here were developed exclusively for predicting energy demand in the residential sector, however the models LEAP and MAED-2 can also determine the energy demand in other sectors (e.g. industrial, transportation and services). This has the evident advantage of providing modelers with the possibility of conducting broader studies of the impact of energy policy measures across several sectors. The disadvantage of these models is that they typically do not have a high level of detail, which means that the influence of certain building related measures such as an improvement of the wall insulation or window glazing (e.g. a change from single to double glazing) cannot be modeled. Here the advantage of models that are focused on the residential sector only such as BREHOMES, CDEM, CREEM, and REEPS shows, as such influences can be modeled and hence the overall impact of improvement measures be detected.

44

A. Martinez Soto, M.F. Jentsch / Energy and Buildings 128 (2016) 38–55

Depending on the amount of data used for the modelling, models may have different levels of disaggregation [8]. In this work one of the models with the highest level of disaggregation is CHM which disaggregates the English housing stock into 14,951 representative dwellings, and each of these cases represents a determined quantity of dwellings, which, added together, represent the total number of dwellings in England (22.8 million in 2011 [61]). Among the models with a moderate level of disaggregation is CDEM which disaggregates the English housing stock into 47 house archetypes defined by built form and age group. The model with the lowest level of disaggregation is LEAP. LEAP disaggregates the energy use in the housing sector according the number of dwellings, average energy end–use per year (space heating, hot water provision, lighting and appliances), energy source and the efficiency of the energy conversion equipment. Overall, it was observed that the developers of residential energy models disaggregate the housing stock using representative housing units which vary in denomination and their selection criteria. For example, CREEM disaggregates the Canadian housing stock into 16 “house archetypes” defined by regional location and age group, whilst BREHOMES disaggregates the UK housing stock into over 1000 “categories”, again defined by age group, built form, tenure and ownership of central heating. By contrast, BSM disaggregates the German housing stock into “buildings types”, such as single family houses (detached houses), two family houses (semi-detached houses) and small and large multi-family houses (including high-rise buildings) which are further divided into eight age groups and the two parts of the country (the former East and West Germany). As highlighted in Table 2, in this paper the term “house archetypes” is used as a collective term for the various dwelling classification systems because their purpose is the same regardless of the exact selection criteria. 3.5. Modelling energy-environment interactions Since the 1990s, a number of models were developed for assessing energy-environment interactions with respect to greenhouse gas emissions [58], demonstrating that there is a linear relationship between the factors that determine energy use in the residential sector and the related carbon dioxide emissions [17,25,52]. These models, which are also shown in Table 2, offer modelers the ability to quantify and visualize the CO2 emissions related to residential energy use. However, it should be noted that these models require additional emission factor data related to the energy sources used in the residential sector, including exact shares of the fuels which make up the energy matrix. 4. Methodology for analyzing the transferability, accuracy and sensitivity of energy models In this section the methodology and parameters used to determine the transferability, accuracy, and sensitivity of the twelve energy models given in Table 1 are explained. 4.1. Transferability For the purpose of this paper model transferability is considered as the ability of a model to quantify the energy use/demand in a country other than that for which it was originally developed. As shown in Table 1 the majority of the energy models investigated here have been designed for use in a specific country within the particular reference framework of this country. According to Bhattacharyya and Timilsina [58] the transferability of energy models is fraught with difficulties which hinder the developed of a wider modelling tool that would be universally applicable. However, to date, a detailed analysis of these difficulties

for model transferability is not readily available. Nevertheless, the transferability of existing energy models would offer the opportunity to avoid the need for developing a new model for each country to be investigated. Furthermore it would allow policy makers to reuse proven/validated existing energy modelling methodologies. To verify the transferability of the models highlighted in Table 1 two criteria were used: (a) the general availability of the required input data and (b) the flexibility of the algorithms to be modified/adapted to represent the framework conditions of a different country. The results of this analysis in conjunction with the proposed adjustments and amendments to the original energy models’ algorithms allow determining the capacity of an energy model to quantify the energy use/demand in a country other than that for which they were created. 4.2. Accuracy The accuracy of the eight energy models assessed here is analyzed in comparison to national statistical energy data for the German residential sector, with the term “accuracy” referring to “the closeness of results obtained by the models to the ‘true’ value”. In this case the “true” value corresponds to the energy data provided by the Federal Ministry of Economic Affairs and Energy (BMWi—Bundesministerium für Wirtschaft und Energie) [40]. This statistical energy data for the German residential sector is determined by the AGEB (AG Energiebilanzen e.V.) and published annually by the BMWi [62]. It is based on empirical data from statistical surveys carried out by DESTATIS (Statistisches Bundesamt) with energy data from energy supply companies [63]. These data are complemented by data from representative surveys of the energy use of private households such as the study: “Energy consumption assessment of private households within the years 2011–2013” (Erhebung des Energieverbrauchs der privaten Haushalte für die Jahre 2011 bis 2013) [64] or the “Sample census 2010” (Mikrozensus 2010) [65]. When these representative surveys are not sufficient to obtain the relevant energy data such as, for example, the subdivision of the final energy use by end use, then bottom-up models are used to derive the missing data (EEFAWohnungsmodell and ZSW-Solarthermiemodell) [62]. To assess the accuracy of the transferable energy models, two data periods are considered: a “reference period” from 1990 to 2000 for comparing model outputs on an annual basis to statistical data for the same period [40], as well as a “forecast period” where annual data from 1990 to 2000 is used to “feed” the models information in order to provide an estimate for the years from 2001 to 2010. All models can use data from the “reference period” for the “forecast period” and the results are again compared to the available national statistical data on the total final energy use in the residential sector [40]. The following analysis parameters are used for the comparison of simulated and statistical data: a) The relative deviation error to determine the difference between the absolute results for the total final energy use/demand in the residential sector of the simulation in relation to the statistical data of the BMWi [40]. b) The mean value of the percentage difference (MPD) which is formed by the absolute difference of the two variables divided by their average to determine the similarity between the simulated and the statistical data. This analysis is conducted separately for the ‘reference period’ and the “forecast period”. This permits both, investigating the ability of the models to determine the final energy use on a year by year basis and to predict the future final energy demand over a time period of 10 years. c) The Pearson’s correlation coefficient r to express the degree of the linear relationship between the total final energy use/demand of the simulated and the statistical data. The Pear-

A. Martinez Soto, M.F. Jentsch / Energy and Buildings 128 (2016) 38–55

45

Table 3 Classification of the Pearson’s correlation coefficient’s values following Zühlke [66]. Correlation

Pearson correlation coefficient r

Interpretation

perfect very strong medium weak none very weak

r=1 0.75 ≤ r < 1 0.50 ≤ r < 0.75 0.00 < r < 0.50 0 −1 ≤ r < 0.00

perfect similarity, identical profiles very similar, significant similarities moderate similarity little similarity no correlation very large differences, opposing curves

son’s correlation coefficient r can assume values between −1 and +1. However, as there is no common classification for the Pearson’s correlation coefficient’s values, the designations shown in Table 3, which are based on Zühlke [66], are used. Furthermore the coefficient of determination r2 is added to assess how well a model explains and predicts future outputs. 4.3. Sensitivity analysis Further to this, a sensitivity analysis testing the robustness of each model is being carried out, also determining model sensitivity to changes in the input parameters. This technique allows to assess which of the input parameters has the greatest impact on the respective model’s output [67]. Here a sensitivity analysis with a one-at-a-time (OAT) approach is used, meaning that the input parameters of the model (kj ) will be individually changed by kj while the other parameters (except the output parameters) remain constant. For each change in the input parameters (kj ), the model is run and the new output variables are used to calculate a sensitivity coefficient ∂yi /∂kj [51] where yi is the ith output variable and kj is jth input parameter. The calculation of the sensitivity coefficient (Si ) is carried out using the following equation: Si =

yi (kj + kj ) − yi (kj − kj ) ∂yi ≈ i = 1, . . ., n and j=1,. . ., m 2kj ∂kj (8)

To be able to compare sensitivity coefficients based on input parameters with different units a normalized sensitivity coefficient (NSC) is then calculated. The normalized sensitivity coefficients Si,j determined according to Eq. (9) then represent the variations in the output variables given a defined percentage change in the input parameters. In this study, variations of ±1% in the input parameters of the model (kj ) are used. Si,j

kj ∂yi = yi ∂kj

i = 1, . . ., n and j = 1, . . ., m

(9)

A sensitivity analysis is deemed essential for the development of any model [68]. In this work, the normalized sensitivity coefficients of the models rather than the sensitivity coefficients are compared in order to avoid inconsistencies in the results. 5. Results 5.1. Transferability Following the criteria mentioned in Section 4, it was found that all statistical models were transferable, however not all building physics models. CREEM, DECM, and CHM fall within the category of non-transferable models. The main reason that these models are non-transferable is because the core calculation engines which these models use (HOT2000, SAP-2005 and SAP-2009 respectively) do not allow the application of data related to the building characteristics or the climates of countries other than those for which

the models were created (Canada or the UK). This ultimately renders it impossible to depict the situation in a different country. The core calculation engine for REEPS: DOE-2 [69] makes this model a further non-transferable model because it requires building data such as the length, width, and orientation of the buildings which is not available for the representative buildings of the German housing stock [70]. Furthermore, as would be expected, the collection of data for more detailed models (i.e. principally the building physics models) was found to be more difficult. It was, however, possible to compile the necessary data to perform simulations with all the eight models that were defined as “transferable”. The data sources for this input data are given in Table 4. Overall, the model outputs regarding the energy use in the “reference period” and energy demand in the “forecast period” in the German residential sector were obtained by re-engineering the models using the methodologies and approaches proposed by the respective model’s authors. However, further adjustments were made where required in order to reflect the particularities of the German housing stock and to facilitate comparison of the results. A summary of these adjustments/changes is shown in Table 5. The remaining two transferable models BSM and MAED-2 not listed in Table 5 did not require any adjustments. Looking at the overall modelling results presented in Table 6 it can be concluded that the eight models used here (REM, BSM, MAED-2, FfE-Gebäudemodell, LEAP, ECCABS, CDEM, BREHOMES) have the capacity to quantify the energy use/demand in a country other than that for which they were originally created. However, as detailed in the following section LEAP and BREHOMES require adjustments in order to improve accuracy as well as to ensure transferability. 5.2. Accuracy The modelling results are presented separately in Figs. 2 and 3 for the statistical models and the building physics models respectively in an attempt to determine the correlation between the degree of disaggregation of the data and the accuracy of the results. Table 6 gives an overview of the accuracy of the model results with respect to the final energy use in the residential sector in Germany in relation to the statistical data given by the BMWi [40], following the parameters defined in Section 4. As a general tendency it is noted that the statistical models have a good ability to determine the energy use in the “reference period” with a mean value for the percentage difference between 5% and 12% compared to the BMWi data (Table 6). However, the ability of these models to predict the energy demand in the “forecast period” in some cases diverges substantially from the BMWi data, especially for LEAP—Version A which has a mean value for the percentage difference of 27% in this period. For this first simulation with LEAP the original software functions were used, i.e. Logistic Forecast and Interpolate functions with the best fit to the development in the “reference period” for the number of dwellings and the energy use for different tasks (space heating, hot water provision, appliances and lighting). It was found that the shortcomings in the prediction

46

A. Martinez Soto, M.F. Jentsch / Energy and Buildings 128 (2016) 38–55

Table 4 Model input parameters and the associated data sources for the German housing stock. Input parameter

Data source

Models requiring this input

Number of dwellings with disaggregated data for new build (addition and replacement) and demolition

[71–78]

BREHOMES, MAED-2, CDEM, REM, FfE-Gebäudemodell, LEAP,BSM, ECCABS,

House archetypes (32 types for the German residential sector disaggregated into four building types (detached family houses, semi-detached family houses, small and large multi-family houses) and the age group (before 1861, 1861–1919, 1919–1949, 1949–1958, 1958–1969, 1969–1979, 1979–1984, 1984–1990)

[78–81]

BREHOMES, CDEM, REM, FfE-Gebäudemodell, BSM, ECCABS, MAED-2

Construction and building component data (U-values, surface area, building volume, air infiltration rate)

[78–85]

BREHOMES, CDEM, ECCABS,

Population

[86,87]

BREHOMES, MAED-2

Persons per dwelling

[71]

MAED-2

Climate data (degree days, external air temperature, solar radiation)

[88,89]

BREHOMES, MAED-2, CDEM, ECCABS,

Equipment and appliances data (tenures, size, efficiency of heating and hot water systems, insulation of boilers)

[90–102]

BREHOMES, CDEM, REM, FfE-Gebäudemodell, MAED-2, LEAP, ECCABS

Reference scenario based on S-curves of the uptake of individual energy related products (insulation of building components, windows with double glazing, electric cookers, proportion of lights with low energy lamps)

[80,97,98,103,104]

BREHOMES, CDEM

Average size (m2 ) and average energy use (electric appliances, space heating, hot water provision, cooking and air conditioning) of dwellings by type

[48,56,71,75,78,81,105–107]

MAED-2, REM, FfE-Gebäudemodell, LEAP, BSM

Energy resources (heating oil, gas, coal, electricity, renewables) and CO2 emission factors

[48,71,108,109]

BREHOMES, CDEM, MAED-2, LEAP, REM, ECCABS

Fig. 2. Comparison of national statistics on the total final energy use in the residential sector in Germany [40] with the modeled energy demand of the statistical models REM, BSM, MAED-2, FfE-Gebäudemodell, LEAP—Version A (considering the original projection of the trend in energy use for space heating as given in the model) and LEAP—Version B (considering an adjusted projection of the trend in energy use for space heating to account for the application of the German Energy Saving Regulation EnEV [110]).

of the energy demand with LEAP—Version A are due to changes in the trend of the input parameters which the model fails to consider. The error with the LEAP—Version A simulation is essentially related to the fact that the two most influential input parameters of this model (the number of dwellings and the average energy use for space heating) faced changes in their trend. Whilst the increase in

the number of dwellings was fairly linear until 2000 with an average growth rate 1.3% per year, this rate decreased to 0.5% from 2001 to 2010 [75]. Further to this the average final energy use for space heating changed in trend from 2002 onwards [71]. Such issues with predictions based on LEAP, where the energy demand is a projection of the historical energy use, were also observed in other studies

A. Martinez Soto, M.F. Jentsch / Energy and Buildings 128 (2016) 38–55

47

Table 5 Summary of the adjustments and changes required to the original energy models for modelling the energy demand in the German residential sector. Model

Adjustments/changes

Reason for adjustment/change

LEAP

(a) Correction of the trend in energy use related to space heating.

(a) The trend in the German residential sector energy use related to space heating displayed a clear change since 2001.

FfE-Gebäudemodell

(a) Addition of a quantification for the total amount of floor space and the number of buildings by type and age group (adapted to the age groups presented in Table 4). (b) Addition of efficiency data for the heating/hot water systems used for the period of investigation (1990–2010). (c) Inclusion of statistical data for the final energy use for cooking, appliances and lighting.a (d) Adjustment of the development in the number of dwellings.

(a) A detailed quantification of the total amount of floor space and the number of buildings by type and age group was not available. (b) Data for the development of the efficiency of the heating/hot water systems) are not considered in the original FfE-Gebäudemodell. (c) The original FfE-Gebäudemodell does not consider energy use for cooking, appliances and lighting. (d) No annual breakdown was available for the number of dwellings for the period of investigation (1990–2010).

REM

(a) Inclusion of statistical data for the final energy use for hot water provision, cooking, appliances and lighting.a (b) Adjustment of the development in the number of buildings and their specific energy demand.

(a) REM does not consider energy use for cooking, appliances and lighting.

(a) Adjustment of the building characteristics and the number of house archetypes.

(a–c) The original data for the climate, building characteristics and the number of house archetypes are for the Swedish residential sector.

ECCABS

(b) The original data regarding the number of buildings and their specific energy demand are for the province of North Karelia (Finland).

(b) Replacement of the climate data to represent Germany (temperature, solar radiation). (c) Adjustment of the development in the number of dwellings. (a–e) The original data for the climate, building characteristics and the number of house archetypes, the ratio between the number of persons and living space per dwelling, the air exchange rate and the heating temperature are for the UK/English residential sector.

(a) Adjustment of the building characteristics and the number of house archetypes.

BREHOMES/CDEM

(b) Replacement of the climate data to represent Germany (temperature, solar radiation). (c) Change in the ratio between the number of persons and the living space per dwelling. (d) Adjustment of the air exchange rate and internal heating temperature. (e) Adjustment of the development in the number of dwellings. a It needs to be noted that the inclusion of the statistical data was made in order to facilitate comparison with the outputs for final energy use/demand from other models. However, this also means that the FfE-Gebäudemodell and REM cannot be used for predicting the future energy demand of a country without further adjustments.

Table 6 Results for the accuracy of the transferable residential energy models according to the parameters defined in section 4 for the case of Germany. Statistical models

Building physics models

REM

BSM

MAED-2

FfE-Gebäudemodell

LEAP A

LEAP B

ECCABS

CDEM

BREHOMES A

BREHOMES B

Reference period Relative deviation error (%) Mean value of the percentage difference (%) Pearson’s correlation coefficient r Coefficient of determination r2

≤15 6 −0.14 0.02

≤20 12 0.80 0.64

≤10 7 0.90 0.81

≤14 7 −0.74 0.55

≤ 13 5 0.62 0.38

≤13 5 0.62 0.38

≤16 8 0.57 0.32

≤17 9 −0.13 0.02

≤24 15 −0.53 0.28

≤16 8 0.34 0.12

Forecast period Relative deviation error (%) Mean value of the percentage difference (%) Pearson’s correlation coefficient r Coefficient of determination r2

≤18 6 −0.74 0.55

≤19 11 −0.57 0.32

≤26 9 −0.55 0.30

≤16 6 −0.69 0.48

≤50 27 −0.67 0.45

≤34 16 −0.67 0.45

≤14 5 −0.67 0.45

≤14 7 −0.59 0.35

≤26 23 0.75 0.56

≤14 6 −0.62 0.38

[111–115]. It is clearly shown that when trends change, there is an imminent error in this type of forecasting where historical trends are used for the modelling.

In the second simulation with LEAP shown in Fig. 2, i.e. with LEAP—Version B, a decrease in the energy demand was considered in the modelling as a reduction in the specific energy demand for new buildings in order to reflect the requirements of the Energy

48

A. Martinez Soto, M.F. Jentsch / Energy and Buildings 128 (2016) 38–55

Saving Regulation EnEV [110]. This results in an improved accuracy in terms of the mean value of the percentage difference (16%). However, it needs to be highlighted that this implies that a user of this model needs to take an informed decision as per the likely changes in the input parameters in order to obtain reliable forecasting results. An important aspect observed in the modelling is that, even though there were thermal insulation regulations in place since 1976 which aimed at reducing the energy demand for space heating, the energy use for space heating clearly increased in the “reference period” (1990–2000). As these regulations should, in principle, over time result in a decrease in energy use for space heating, the question is when this decrease would be detectable in the national statistical data for the residential sector. The answer to this could then be a contribution for improving the accuracy of energy prediction models. To elaborate predictions, modelers generally develop a series of scenarios which represent possible development pathways. However, it is often observed that the measures adopted in such scenarios have an immediate effect on the forecast [22,116]. In the case presented here this observation however contrasts with the statistical data, where it is observed that the impact of energy saving measures needs time to feed through to the results. For example, if the data from 1990 to 2010 for overall final energy use in the German residential sector in Fig. 2 is considered, it can be assumed that legislation to reduce the energy demand by a reduced space heating demand in new buildings only had a noticeable effect on the overall final energy use trend since about 2001. This is because the diminution of final energy use for space heating of recently constructed buildings can, for example, be offset by an increase in the number of dwellings, which ultimately contributes to an overall increase in energy demand of the country. This was the case in the German residential sector. Within the group of the statistical models given in Fig. 2, the module to predict the number of buildings in the model BSM needs to highlighted. Here the total building stock in the year t + 1 is obtained from the stock in year t, considering both the number of demolished buildings in year t and the number of new buildings constructed in year t. The number of demolished buildings and the forecast for the number of new buildings are determined by an equation which considers the construction of new buildings, looking at both addition and replacement, as well as building demolition. It is based on studies related to the German building stock [72,76,77,87]. As a result BSM was the model which most accurately predicted the number of dwellings with a mean value for the percentage difference of this parameter of 2% in the “forecast period”. The remaining three statistical models shown in Fig. 2 (REM, FfE-Gebäudemodell and MAED-2) display a high level of accuracy, with a mean value for the percentage difference between 6% and 7% for the reference period and 6% and 9% for the forecast period (Table 6). The MAED-2 methodology is notable as the model results have a very strong correlation to the BMWi data in the “reference period”. The reason for this is that MAED-2 considers degree day data and can, according to the methodology given in Section 4, use the statistical values for the degree days in the “reference period”. Therefore, a clear correlation is observed as the degree days have a strong influence on the final energy use in the residential sector. By contrast, in the “forecast period” a mean value was used for the degree days which, as would be expected, led to a loss of the correlation. As highlighted in Table 5 the energy model REM can only quantify the final energy use related to space heating and the FfEGebäudemodell only the final energy use related to space heating and hot water provision. The final energy use related to lighting, appliances, cooking and hot water provision (REM only) was, there-

fore, added to these two models using statistical data provided by the Federal Environment Agency (Umweltbundesamt) [48]. As indicated in Table 5 this was done in order to allow comparison with the outputs of the other models. This approach was considered as appropriate since the final energy use in the German residential sector largely consists of space heating and hot water provision (>86%) [109]. It should be noted however, that the use of statistical data to complement the results obtained by the models FfE-Gebäudemodell and REM to some extent improves the accuracy of the results. For example, if the final energy use results for space heating quantified by the REM model for the period 1990–2010 are compared with data from the Federal Environment Agency, one observes that the relative deviation error reaches up to 21%. At the same time the mean value of the percentage difference equals 7%. When the model however includes the statistical data for the final energy use for hot water provision, cooking, appliances and lighting, then the relative deviation error is reduced by 3%-points and the mean value of the percentage difference is likewise reduced by 1%-point. The same is true for the results of the FfE-Gebäudemodell model, where the relative deviation error is reduced by 2%-points and the mean value of the percentage difference by 1%-point. As can be seen in Fig. 3, the bottom-up building physics models investigated in this study in general deliver results with a low error margin for the case study country Germany, with a mean value for the percentage difference between 5% and 9% (Table 6). An exception to this is the simulation with BREHOMES—Version A which followed the original method by Shorrock and Dunster [22] who suggest that the total number of dwellings is based on projections for the population and the mean household size. The simulation with BREHOMES—Version A was based on these two criteria as per the equation in BREDEM, i.e. the core calculation engine of BREHOMES which compares the number of occupants to the total floor area of the dwellings. Shorrock and Dunster [22] however already warned that the proposed relationship between the number of occupants and the floor area of dwellings is only a rough indicator and that a large variation could arise in practice. Accordingly, it was observed that, for the case of Germany, the results obtained for the number of dwellings proved to be lower than the actual number and hence the results for the final energy use in the residential sector were also lower than those presented by the DESTATIS [75]. Therefore, it can be concluded that the original equation and calculation criteria for the UK are not transferable to other countries. Furthermore, it was found that the BREDEM calculation routine assumes a constant ratio between the number of occupants and floor area (30.5 m2 / Person). However, this relationship is subject to changes as highlighted by the UBA [48]. For example, in Germany this ratio was 11.7 m2 per person in 1950 and 34.7 m2 per person in 1990 respectively (calculation based on [79,117]). To overcome the shortfalls with BREHOMES—Version A a new simulation (BREHOMES—Version B) was performed, this time considering a projection of the number of dwellings based on growth rates obtained for Germany in the “reference period”. With this alteration the results improve substantially (from 23% to 6% for the mean value of the percentage difference). BREHOMES and CDEM are both BREDEM-based models. Therefore, as the same data sources were used to calculate the final energy use in the German residential sector during the “reference period”, principally the same results would be expected for these two models. These models are, however, differentiated according to the level of disaggregation of the housing stock. CDEM disaggregates the English housing stock into 47 house archetypes and BREHOMES disaggregates the UK housing stock into over 1000 categories to determine the final energy use of a base year (Table 2). To replicate the difference between both methodologies, two different levels of disaggregation for the base year (1990) were used for calculating the final energy use for the case of Germany. For

A. Martinez Soto, M.F. Jentsch / Energy and Buildings 128 (2016) 38–55

49

Fig. 3. Comparison of national statistics on the total final energy use in the residential sector in Germany [40] with the modeled energy demand of the building physics models ECCABS, CDEM, BREHOMES—Version A (considering the original BREDEM algorithms for the number of dwellings) BREHOMES—Version B (considering the reference period growth rates for calculating the number dwellings).

the simulation with BREHOMES the German housing stock was disaggregated into 32 dwelling types following 2007 data of the IWU for the housing stock [79]. By contrast, for the simulation with CDEM 16 dwelling types were used which were selected according to the built form (four types) and the years with modifications to the energy saving regulations (WärmeschutzV) [118] (1977, 1984, 1995), resulting in four building age groups. This permits an investigation into whether a more detailed disaggregation of the dwelling types provides more accurate results. To predict the energy demand within the “forecast period” the BREHOMES model uses an average dwelling in the BREDEM calculation engine which is based on average as well as fully-disaggregated data for the building stock and S-curves [22]. By contrast, the results of CDEM are based on the prediction of the energy demand for each individual dwelling type as calculated by BREDEM [17]. The difference between both results is, however, marginal. They display the same relative deviation error and a 1%-point difference for the mean value of the percentage difference (Table 6). This appears to indicate that it may be sufficient to use the less complex methodology of BREHOMES, because this facilitates data input for the calculation and takes less time to model. Yet it should be noted that the methodologies used for CDEM allow for modelling decisions regarding energy conservation strategies that are more focused on individual measures such as the renovation of old buildings or improving the thermal insulation of a specific age group of buildings. Overall, Table 6 demonstrates that in terms of accuracy in quantifying energy use within the “reference period”, both, the statistical and the building physics models delivered similar results with a mean value of the percentage difference between 5% and 15%. The results show, that only three models (REM, FfE-Gebäudemodell and CDEM) have a very weak correlation r, while, according to the definition given in Table 3, the other models have a medium or a very strong correlation r. The model with the highest coefficient of determination r2 is a statistical model (MAED-2). This is due to the fact that, in the “reference period”, MAED-2 can consider the real values for the degree days which increases the correlation between the results of the model and the BMWi data where the underlying modelling also considers the climate for each particular year. Furthermore, the methodology used by AGEB (AG Energiebilanzen e.V.) [62] for calculating the final energy use in the residential sector is very similar to the methodology of MAED-2. Therefore, similar

results would be expected for the “reference period”. However, conversely to MEAD-2 the AGEB methodology cannot be applied for predicting the future energy demand in the residential sector [62]. Interestingly, as shown by the mean values of the percentage difference in Table 6, the results of the three relevant building physics models (ECCABS, CDEM, BREHOMES—Version B) increase in accuracy for the “forecast period” as compared to the “reference period”, whereas for the statistical models no such trend can be observed. This also shows in Figs. 2 and 3 where the statistical models display a continuous increase in energy demand over the “forecast period”, whilst the building physics models only depict a small increasing trend. This is basically due to the fact that the statistical models use trend projections for predicting the future energy demand. As the most influential parameters, i.e. the number of dwellings and the specific energy demand continuously increased over the “reference period” this is also projected to the “forecast period”, resulting in a general overestimation of the “future” energy demand. By contrast, the building physics models essentially predict the future energy demand by considering a trend projection in the number of dwellings as well as based on S-curves the impact of energy saving measures such as the improvement of the efficiency of heating and hot water systems, the insulation of building components, the proportion of windows with double glazing or the proportion of lights with low energy lamps. Whilst the former produces an increase in the “future” energy demand, the latter results in a decrease. Finally, the combined effect of all input parameters is a slight increase of the overall energy demand in all the investigated building physics models, which as a trend appears to better reflect the general tendency of the BMWi data [40] over the years 2001–2010 than the statistical models. Table 6 also shows that, in the “reference period”, the Pearson’s correlation coefficient r was positive for the results of most models. However, in the “forecast period”, the values of the Pearson’s correlation coefficient were negative for the majority of models. This is due to the fact that most models predicted an increase in energy demand, whilst the trend of the BMWi data indicates a reduction in energy use since 2001.

5.3. Sensitivity analysis In 2010 Firth et al. [17] presented the first residential sector energy model that included a sensitivity analysis at the model

50

A. Martinez Soto, M.F. Jentsch / Energy and Buildings 128 (2016) 38–55

development stage: CDEM, which is a BREDEM based model that uses housing stock data based on the English House Condition Survey 2001 (EHCS) [119]. The results of this sensitivity analysis show that the CDEM model has the highest sensitivity (Si,j = 1.55) to mean internal temperature followed by external air temperature (Si,j = −0.58) and heating system efficiency (Si,j = −0.58) [17]. Here, the negative values indicate that the input parameter is inversely proportional to the output parameter (CO2 emissions), i.e. that an increase in the input parameter will lead to a reduction in CO2 emissions. In 2011 Cheng and Steemers [25] presented the DECM residential sector energy model which also included a sensitivity analysis. DECM is a model based on the UK Government Standard Assessment Procedure (SAP-2005) [33], which itself is a simplified version of BREDEM-12 [30]. In addition to this DCEM uses housing stock data derived from the English House Condition Survey 2007 (EHCS) [120]. The results of the sensitivity analysis refer to the average CO2 emissions of English dwellings and again highlight that the model has the highest sensitivity (Si,j = 1.55) to mean internal temperature, however in this case followed by total floor area (Si,j = 0.77) and external air temperature (Si,j = −0.61) [25]. Although both models were found to be most sensitive to mean internal temperature, with the analyses even yielding identical sensitivity coefficients for this parameter, they do not agree on the influence of other input parameters. According to Cheng and Steemers [25] this inconsistency can be attributed to differences in the model assumptions as well as the different data sources used for the input data. In 2013 Hughes et al. [51] presented the Cambridge Housing Model (CHM), also including a sensitivity analysis. Like DECM CHM uses English housing stock data, in this case the data given in the English Housing Survey 2011 [121]. The results of the sensitivity analysis show that the model again has the highest sensitivity (Si,j = 1.54) to mean internal temperature, in this case followed by heating system efficiency (Si,j = −0.66) and external air temperature (Si,j = −0.59) [51]. Hughes et al. [51] compared their results to those of Firth et al. [17] and emphasized the general consistency of the results. However, they did not discuss the difference in influence of specific input parameters, in particular the second most influential parameters “heating system” and “floor area”. In 2013 McKenna et al. [24] proposed the residential energy model BSM which was developed to predict the energy demand in the German residential sector using the housing stock data as provided by BBSR [72], DESTATIS [73–75], and Bayern LB [76]. The results of the sensitivity analysis that was conducted by McKenna et al. [24] show that the BSM model has the highest sensitivity (Si,j = 0.54) to the level of refurbishment followed by the building replacement rate (Si,j = 0.2) and the renovation probability (Si,j = 0.1). When these sensitivity analysis results are compared to those of the three models CDEM, DCEM and CHM discussed above it becomes clear, that not only differences in the values for the sensitivity coefficient are observed, but that the key model parameters may also diverge depending on the overall modelling approach. Therefore, in order to facilitate comparison, the values for the normalized sensitivity coefficients (Si,j ) were calculated for the German residential sector and ranked in descending order for the eight “transferable” models as shown in Figs. 4 and 5 for the examples of the CDEM and the MAED-2 model respectively. The calculation followed the methodology given in Section 4. As would be expected, it was found that there is a vast variety of input parameters that are difficult to associate and compare. The normalized sensitivity coefficient is directly (for positive Si,j ) or indirectly (for negative Si,j ) proportional to a 1% variation in the respective input parameter. For example, as the “population” input parameter has the greatest influence on the sensitivity (Si,j = 1.98) of the MAED-2 model (Fig. 5), this means that a 1% rise in the population results in a 1.98% increase in the final energy demand. On the other hand, for the same model (MAED-2) a 1% rise in the parame-

ter “persons per house” yields a 1.73% reduction in the final energy demand (Si,j = − 1.73). Fig. 4 compares the results of the sensitivity analyses with regards to predicting CO2 emmisions when using the CDEM model for the UK as undertaken by Firth et al. [17] and the German residential sector investigated in this paper. It can be clearly seen that the most influential parameters for the same model vary with different boundary conditions, i.e. when analyzing different countries. Furthermore, it can be seen in Fig. 4 that, in the sensitivity analysis of Firth et al. [17], the influence of the total number of dwellings was not quantified. This is because Firth et al. [17] chose to analyze the effects of input parameters on the CO2 emissions of a single average dwelling. For the case of Germany however, the number of dwellings was considered since the aim was to analyze the effect of the input parameters on the CO2 emissions of the entire residential sector. If the sensitivity analysis results are compared to those from a different model, a different ranking with respect to the individual parameters’ influence is obtained. This is highlighted by Fig. 5 for the sensitivity analysis results of MAED-2 for the German residential sector in comparison to those of CDEM shown in Fig. 4. It can be clearly seen that the input parameters follow different philosophies which makes it difficult to directly compare the results. For CDEM the five most influential input parameters are: the number of dwellings, the heating demand temperature, the external temperature, the infiltration rate and the length of the daily heating period (Fig. 4). For MAED-2 the five most influential input parameters are: the population, the number of persons per dwelling, the degree days, the floor area by dwelling type and the specific heat loss rate by dwelling type (Fig. 5). A comparison of all the eight models becomes even more complex, because the diversity and number of parameters further increase. Nevertheless, it is observed, that, when ordering the normalized sensitivity coefficients of all individual models, the following sequence for the most influential input parameters in decreasing order is common to all models: the number of dwellings and the specific energy demand for (a) space heating, (b) hot water provision and (c) appliances and lighting. However, these four input parameters are often not directly inputs to the respective models. For example, if the two most influential input parameters are population and the number of persons per dwelling, such as is the case in MAED-2, then these input parameters are used to determine the number of dwellings, i.e. the most influential common parameter across all models. Similarly, in order to determine the specific energy demand for space heating several input parameters may be used. For example, in the case of CDEM, the four most influential parameters after the number of dwellings, i.e. the heating demand temperature, external temperature, infiltration rate, and the length of the daily heating period, are used to determine the specific energy demand for space heating. Overall, this shows that there is an intrinsic accordance between the respective energy models and that the difference lies in the level of detail of the input parameters used to determine the energy demand. Based on this observation Table 7 attempts to group input parameters according to both their degree of influence and the level of detail available for the respective input parameters. As can be seen in Table 7 two levels of detail were determined in relation to the models’ level of disaggregation. Models with a low level of disaggregation (e.g. LEAP, MAED-2, REM) use input parameters without a large breakdown of detail, such as for example the total number of dwellings and the specific energy demand for different uses (space heating, hot water provision, appliances and lighting). Models with a high level of disaggregation (e.g. CDEM, BREHOMES) use input parameters with a high breakdown of detail, e.g. heating demand temperature, U-value and surface area of building components or infiltration rate.

A. Martinez Soto, M.F. Jentsch / Energy and Buildings 128 (2016) 38–55

51

Fig. 4. Comparison of the normalized sensitivity coefficients obtained with CDEM for the German and the UK residential sector (Data for Germany own calculation and for the UK based on [17]).

Table 7 General grouping of key input parameters for residential energy models according to their degree of influence and the level of detail in the input data as determined for the case of the German residential sector. Degree of influence

Primary level of detail (models with a low level of disaggregation)

Secondary level of detail (models with a high level of disaggregation)

Strong influence (a 1% variation in this parameter causes a change greater than 1% of the total final energy demand: Si,j > 1)

Number of dwellings

– Population – Persons per dwelling – Persons per dwelling area

Specific energy demand for space heating

– – – – – –

Specific energy demand for hot water provision

– Boiler Efficiency – Persons per dwelling – Persons per dwelling area

Specific energy demand for appliances and lighting

– Efficiency of appliances and lighting – Persons per dwelling area

Medium influence (a 1% variation in this parameter causes a change between 0.1% and 1% of the total final energy demand: 1 ≥ Si,j ≥ 0.1) Low influence (a 1% variation in this parameter causes a change of less than 0.1% of the total final energy demand: 0.1 > Si,j )

However, it must be emphasized that the ordering of the input parameters presented in Table 7 was carried out on the basis of data from the German residential sector and, therefore, may be different for other countries. This could particularly be the case for countries with a climate that requires space cooling rather than space heating.

Heating demand temperature Degree days External air temperature Boiler efficiency Infiltration rate U-value and surface area of building components

6. Discussion In this work the transferability and accuracy of twelve residential sector bottom-up energy models was analyzed, taking Germany as a case study example. This also included a sensitivity analysis. Four of the investigated energy models (CREEM, DECM, CHM, and REEPS) were found to be non-transferable. This is basically due to

52

A. Martinez Soto, M.F. Jentsch / Energy and Buildings 128 (2016) 38–55

Fig. 5. Normalized sensitivity coefficients obtained with MAED-2 for the German residential sector (SFH—single family house, MFH—multi family house).

the core calculation engines which these models use, i.e. HOT-2000 [29], SAP-2005 [33], SAP-2009 [34] and DOE-2 [69] respectively. Whilst HOT-2000, SAP-2005 and SAP-2009 do not allow using climate data from other countries, DOE-2 requires detailed building data such as the length, width and orientation of the buildings which are not readily available for the building types representative of the German housing stock. In terms of the accuracy of results, no significant differences were found between the bottom-up building physics models and the bottom-up statistical models. This is highlighted in Fig. 6 which shows the percentile of the relative deviation error of the eight transferable residential energy models according model group and period of investigation. However, Fig. 6 also demonstrates that the bottom-up building physics models investigated here were found to slightly increase in accuracy in the “forecast period” as compared to the “reference period”, whereas no such trend was observed for the statistical models. As can be seen in Table 6, when disregarding the versions A of LEAP and BREHOMES the bottom-up building physics models, on average, had a mean value of the percentage difference of 8% for the “reference period” and 6% for the “forecast period” and the bottom-up statistical models displayed values of, on average, 7% for the “reference period” and 10% for the “forecast period” respectively. Additionally, it was observed that, although most models were able to determine the trend of the energy use in the “reference period” (see Pearson’s correlation coefficient in Table 6), none of the models was able to predict the trend of the energy demand in the “forecast period”. It should be noted that the accuracy of the results depends greatly on the quality of the input data. This is especially the case for data related to the specific energy demand for space heating and hot water provision, which, besides the number of dwellings, were

Fig. 6. Percentile of the relative deviation error of the eight transferable residential energy models (REM, BSM, MAED-2, FfE-Gebäudemodell, LEAP—Version B, ECCABS, CDEM, BREHOMES—Version B), separated into statistical and building physics models (see Table 6 for the data of the individual models).

found to be the most influential model input parameters (Table 7). Only if there are representative surveys or studies available for the existing housing stock, the statistical models can provide results as accurate as those of the building physics models [122]. For the case of Germany, the surveys carried out by IWU [81,83] and DENA [56] appear to be reliable in this respect. Furthermore, it has been

A. Martinez Soto, M.F. Jentsch / Energy and Buildings 128 (2016) 38–55

demonstrated in Fig. 2 and Table 6 that the use of trend projections can be a source of error when predicting the future energy demand with bottom-up statistical models, particularly when there is an imminent probability for a change of the trends. Energy use in the German residential sector is highly influenced by the number of dwellings and the specific energy demand per dwelling for space heating. For example, it was found that the number of dwellings has increased throughout the “reference period” (1990–2000), leading to an overall increase in energy use in the residential sector. This is somewhat surprising as for new dwellings a reduction in energy demand for space heating has already been promoted since 1977 in West Germany and 1976 East Germany respectively through regulations related to thermal insulation and from 1978 onwards regarding the energy efficiency of boilers (West Germany) [118]. Therefore, a reduction in overall final energy use would be expected even with a rising number of buildings due to building replacement and refurbishment. However, there appears to be a change in the trend for the energy use in the German residential sector from 2001 onwards. From this the question arises how to determine the time that is required for energy saving measures to have any impact and how this interacts with other parameters such as an overall growth rate in the number of dwellings. As was shown in the sensitivity analysis in Section 4, if the degree of influence of each input parameter is known, then it is possible to estimate the induced changes in the total final energy demand due to a particular combination of input parameters. This can then be used to inform policy measures aiming at energy conservation. Moreover, it was determined that the magnitude of the normalized sensitivity coefficients depends on the level of detail in the input parameters. The input parameter “Specific energy demand for space heating” for example used in models with a low level of disaggregation has a larger normalized sensitivity coefficient (average Si,j = 1,48 across all models) than the input parameters “Boiler efficiency” and “Infiltration rate” (average Si,j = 1,1 and Si,j = 0.63 respectively across all models) used in models with a higher level of disaggregation.

7. Conclusions Overall, it can be concluded that, in principle, residential energy models using both modelling approaches, bottom-up building physics and bottom-up statistical modelling, can be transferred to a different country and, depending on the quality of the input data, deliver accurate results. Conducting a sensitivity analysis allows the input parameters with a major influence to be identified and, for the case of the energy use in the residential sector, delivers useful information for the development of effective energy saving measures. However, the results of the sensitivity analysis for the case study country also show that the most influential parameters are difficult or impossible to regulate by energy policies (e.g. number of dwellings, persons per dwelling, heating demand temperature, external air temperature). Additionally, it can be concluded that determining the most influential input parameters depends on the model used for modelling the residential energy demand of a country as well as the specific conditions of the country such as for example the underlying climate (heating dominated or cooling dominated). Therefore, if other countries than Germany are to be investigated, further sensitivity analyses appear to be required. Overall, a multi-model analysis as conducted here is considered advantageous in order to gain a better understanding of both the relevance of input parameters for the case study country and possible pathways for the development in energy use. Generally speaking, it was observed that the investigated statistical models are more sensitive to variations of the input

53

parameters than the building physics models. This would imply that the results of the statistical models have a larger level of uncertainty due to their inherent simplifications and their limited amount of input data. Further work appears to be required regarding the individual uncertainties and interactions of all the input parameters, which would allow determining the overall uncertainty of the results. This would then facilitate the development of a new transferable residential energy model on the basis of several existing energy models’ calculation routines. Acknowledgements This work was generated within the framework of a PhD scholarship by the “Comisión Nacional de Investigación Científica y Tecnológica” (CONICYT) of Chile (up to 01/2015) and since 02/2015 as part of a “Thüringer Graduiertenförderung” PhD scholarship granted by the Free State of Thuringia, Germany. References [1] L. Pérez-Lombard, J. Ortiz, C. Pout, A review on buildings energy consumption information, Energy Build. 40 (3) (2008) 394–398. [2] OECD, IEA, Key World Energy Statistics, International Energy Agency, Paris, France, 2014. [3] EEA, Final Energy Consumption by Sector and Fuel (CSI 027/ENER 016), European Environment Agency, 2015. [4] IEA, Worldwide Trends in Energy Use and Efficiency, International Energy Agency, Paris, France, 2008. [5] L. Swan, V.I. Ugursal, Modeling of end-use energy consumption in the residential sector: a review of modeling techniques, Renew. Sustain. Energy Rev. 13 (8) (2009) 1819–1835. [6] R. Saidur, H.H. Masjuki, M.Y. Jamaluddin, An application of energy and exergy analysis in residential sector of Malaysia, Energy Policy 35 (2) (2007) 1050–1063. [7] J. Munksgaard, K.A. Pedersena, M. Wien, Impact of household consumption on CO2 , Energy Econ. 22 (2000) 423–440. [8] M. Kavgic, A. Mavrogianni, D. Mumovic, A. Summerfield, Z. Stevanovic, M. Djurovic-Petrovic, A review of bottom-up building stock models for energy consumption in the residential sector, Build. Environ. 45 (7) (2010) 1683–1697. [9] COM, 20 20 by 2020: Europe’s Climate Change Opportunity—2008: Communication from the Commission to the European Parliament, the Council, the Economic and Social Committee and the Committee of the Regions, Commission of the European Communities, Brussels, 2008. [10] EU, Directive 2009/28/EC of the European Parliament and of the Council of 23 April 2009 on the promotion of the use of energy from renewable sources and amending and subsequently repealing Directives 2001/77/EC and 2003/30/EC, Off. J. Eur. Union 140 (2009) 16–62. [11] EU, Directive 2012/27/EU EC of the european parlament and of the council of 25 October 2012 on energy efficiency, amending directives 2009/125/EC and 2010/30/EU and repealing directives 2004/8/EC and 2006/32/EC, Off. J. Eur. Union 315 (2012) 1–56. [12] C. Delmastro, E. Lavagno, G. Mutani, Chinese residential energy demand: scenarios to 2030 and policies implication, Energy Build. 89 (2015) 49–60. [13] K. Promjiraprawat, P. Winyuchakrit, B. Limmeechokchai, T. Masui, T. Hanaoka, Y. Matsuoka, CO2 mitigation potential and marginal abatement costs in Thai residential and building sectors, Energy Build. 80 (2014) 631–639. [14] J.-P.A. Snäkin, An engineering model for heating energy and emission assessment the case of North Karelia, Finland, Appl. Energ. 67 (2000) 353–381. [15] IAEA, Model for Analysis of Energy Demand (MAED-2) User’s Manual, International Atomic Energy Agency, Vienna, Austria, 2006. [16] R. Corradini, M. Beer, T. Schmid, Energiemodell der Wohngebäude, BWK – Das Energie-Fachmagazin 64 (1/2) (2012) 48–53. [17] S.K. Firth, K.J. Lomas, A.J. Wright, Targeting household energy efficiency measures using sensitivity analysis, Build. Res. Inf. 38 (1) (2010) 25–41. [18] H. Farahbakhsh, V.I. Ugursal, A.S. Fung, A residential end-use energy consumption model for Canada, Energy Res. 22 (13) (1998) 1133–1143. [19] È. Mata, A. Kalagasidis, Calculation of Energy Use in the Swedish Housing – Description of the Building Energy Simulation Model EABS: Energy Assessment of Building Stocks, Chalmers University of Technology, Gothenburg, Sweden, 2009. [20] É. Mata, A.S. Kalagasidis, F. Johnsson, A modelling strategy for energy, carbon, and cost assessments of building stocks, Energy Build. 56 (2013) 100–108. [21] J.G. Koomey, R.E. Brown, R. Richey, F.X. Johnson, A.H. Sanstad, L. Shown, Residential Sector End-use Forecasting with Eprireeps 2.1: Summary Input Assumptions and Results, Lawrence Berkeley Laboratory, Berkley, USA, 1995.

54

A. Martinez Soto, M.F. Jentsch / Energy and Buildings 128 (2016) 38–55

[22] L.D. Shorrock, J.E. Dunster, The physically-based model BREHOMES and its use in deriving scenarios for the energy use and carbon dioxide emissions of the UK housing stock, Energy Policy 25 (12) (1997) 1027–1037. [23] C. Heaps, User Guide for LEAP, 2008, Stockholm Environment Institute−U.S. Center, Somerville, USA, 2010. [24] R. McKenna, E. Merkel, D. Fehrenbach, S. Mehne, W. Fichtner, Energy efficiency in the German residential sector: a bottom-up building-stock-model-based analysis in the context of energy-political targets, Build. Environ. 62 (2013) 77–88. [25] V. Cheng, K. Steemers, Modelling domestic energy consumption at district scale: a tool to support national and local energy policies, Environ. Modell. Software 26 (10) (2011) 1186–1198. [26] Cambridge architectural research, Cambridge Housing Model and User Guide, Department of Energy & Climate Change, 2010. [27] B. Chateau, B. Lapillonne, Energy Demand: Facts and Trends—A Comparative Analysis of Industrialized Countries, Springer-Verlag GmbH, Wienna, Austria, 1982. [28] É. Mata, A. Sasic Kalagasidis, F. Johnsson, Energy usage and technical potential for energy saving measures in the Swedish residential building stock, Energy Policy 55 (2013) 404–414. [29] HBCS, HOT2000, Natural Resources Canada, Ottawa, Canada (2014). [30] B.R. Anderson, P.F. Chapman, N.G. Cutland, C.M. Dickson, G. Henderson, J.H. Henderson, P.J. Iles, L. Kosmina, L.D. Shorrock, BREDEM-12 Model Description 2001 Update, Building Research Establishment (BRE), Watford, UK, 2010. [31] F.X. Johnson, R.E. Brown, J.W. Hanford, A.H. Sanstad, J.G. Koomey, Residential HVAC Data, Assumptions and Methodology for End-Use Forecasting with EPRI-REEPS 2.1, Lawrence Berkeley Laboratory, Berkeley, USA, 1994. [32] R.J. Hwang, F.X. Johnson, R.E. Brown, J.W. Hanford, J.G. Koomey, Residential Appliance Data, Assumptions and Methodology for End-Use Forecasting with EPRI-REEPS 2.1, Lawrence Berkeley Laboratory, Berkley USA, 1994. [33] BRE, SAP 2005, The Government’s Standard Assessment Procedure for Energy Rating of Dwellings, Building Research Establishment, Watford, UK, 2009. [34] BRE, SAP 2009, The Government’s Standard Assessment Procedure for Energy Rating of Dwellings, Building Research Establishment, Watford, UK, 2011. [35] BREE, Australian Energy Statistics, Bureau of Resources and Energy Economics, Canberra, Australia, 2014. [36] OEE, Energy Use Data Handbook, 1990–2010, 11 ed., Office of Energy Efficiency Natural Resources Canada,Otawa, Canada (2013). ˜ de un Modelo de Proyección de Demanda Energética [37] R. O‘Ryan, Diseno Global Nacional de Largo Plazo, Comisión Nacional de Energía, Santiago, Chile, 2008. [38] S. Chen, N. Li, J. Guan, Y. Xie, F. Sun, J. Ni, A statistical method to investigate national energy consumption in the residential building sector of China, Energy Build. 40 (4) (2008) 654–665. [39] MEDDE, Parc des chauffages centraux des résidences principales, en milieu d’année jusqu’en 2009 (2009). [40] BMWi, Zahlen und Fakten Energiedaten Nationale und Internationale Entwicklung, Bundesministerium für Wirtschaft und Technologie, Berlin, Germany, 2015. [41] ENEA, Rapporto annuale efficienza energetica, Agenzia nazionale per le nuove tecnologie, l’energia e lo sviluppo economico sostenibile, Rome, Italy, 2012. [42] ECCJ, Japan Energy Conservation Handbook 2011, Tokyo, Japan (2011). [43] MAPS, Energy Modelling for the Residential Sector A Summary of the SATIM Methodology, Mitigation Action Plans and Scenarios, Cape Town, South Africa, 2013. ˜ [44] IDAE, Análisis del consumo energético del sector residencial en Espana, Instituto para la Diversificación y Ahorro de la Energía, Madrid, Spain, 2011. [45] EIA, Residential Sector Key Indicators and Consumtion, Reference case, Washington, DC (2015). [46] DECC, Statistics Energy Consumption in the UK, Departament of Energy and Climate Change, London, UK, 2014. [47] N. Diefenbach, H. Cischinsky, M. Rodenfels, K.-D. Clausnitzer, Datenbasis Gebäudebestand Datenerhebung zur energetischen Qualität und zu den Modernisierungstrends im deutschen Wohngebäudebestand, Stuttgart, Germany (2011). [48] Öko-Institut, FhG-ISI, Energieeffizienz in Zahlen, Umweltbundesamt, Dessau-Roßlau, Germany, 2011. [49] AGEB, Auswertungstabellen zur Energiebilanz für die Bundesrepublik Deutschland 1990 bis 2012, Arbeitsgemeinschaft Energiebilanzen e.V., Berlin, Germany, 2013. [50] DESTATIS, Energieverbrauch der privaten Haushalte von 1995 bis 2006 leicht gesunken, Statistisches Bundesamt, Wiesbaden, Germany, 2008. [51] M. Hughes, J. Palmer, V. Cheng, D. Shipworth, Sensitivity and uncertainty analysis of England’s housing energy model, Build. Res. Inf. 41 (2) (2013) 156–167. [52] M. Kavgic, D. Mumovic, A. Summerfield, Z. Stevanovic, O. Ecim-Djuric, Uncertainty and modeling energy consumption: sensitivity analysis for a city-scale domestic energy model, Energy Build. 60 (2013) 1–11. [53] H.K. Jacobsen, Integrating the bottom-up and top-down approach to energy-economy modelling: the case of Denmark, Energy Econ. 20 (4) (1998) 443–461.

[54] D. Johnston, A Physically-Based Energy and Carbon Dioxide, Ph.D Thesis, Leeds Metropolitan University, Leeds, UK, 2003. [55] L.G. Swan, V.I. Ugursal, I. Beausoleil-Morrison, Occupant related household energy consumption in Canada: estimation using a bottom-up neural-network technique, Energy Build. 43 (2011) 326–337. [56] DENA, Dena-Gebäudereport Statistiken und Analysen zur Energieeffizienz im Gebäudebestand, Deutsche Energie-Agentur GmbH, Berlin, Germany, 2012. [57] T. Nakata, Energy-economic models and the environment, Prog. Energy Combust. Sci. 30 (4) (2004) 417–475. [58] S.C. Bhattacharyya, G.R. Timilsina, A review of energy system models, Int. J. Energy Sect. Manage. 4 (4) (2010) 494–518. [59] S. Heiple, D.J. Sailor, Using building energy simulation and geospatial modeling techniques to determine high resolution building sector energy consumption profiles, Energy Build. 40 (8) (2008) 1426–1436. [60] A.T. Booth, R. Choudhary, D.J. Spiegelhalter, Handling uncertainty in housing stock models, Build. Environ. 48 (2012) 35–47. [61] M. Hughes, J. Palmer, P. Pope, A Guide to the Cambridge Housing Model, Cambridge Architectural Research Limited, Cambridge, UK, 2013. [62] H.G. Buttermann, T. Baten, F. Musiol, T. Nieder, R. Erler, T. Raabe, Verfahren zur regelmäßigen und aktuellen Ermittlung des Energieverbrauchs in nicht von der amtlichen Statistik erfassten Bereichen, Bundesministerium für Wirtschaft und Technologie (BMWi), Münster Germany, 2015. [63] DESTATIS, Erhebung über Abgabe, Ein- und Ausfuhr von Erdgas und Erdölgas sowie Erlöse der Produzenten, Statisches Bundesamt, Wiesbaden, Germany, 2013. [64] M. Frondel, M. Andor, N. Ritter, S. Sommer, C. Vance, P. Matuschek, U. Müller, Erhebung des Energieverbrauchs der privaten Haushalte für die Jahre 2011–2013, Rheinisch-Westfälisches Institut für Wirtschaftsforschung (RWI) Essen & forsa Gesellschaft für Sozialforschung und statistische Analysen mbH, 2015. [65] DESTATIS, Mikrozensus2010, Qualitätsbericht, Statistisches Bundesamt, Wiesbaden (2011). [66] J.P. Zühlke, Die Verbreitung von Wissen zu Controlling-Instrumenten, Deutscher Universitäts-Verlag, Wiesbaden, Germany, 2007. [67] J. Keirstead, C. Calderon, Capturing spatial effects, technology interactions, and uncertainty in urban energy and carbon models: retrofitting newcastle as a case-study, Energy Policy 46 (2012) 253–267. [68] A. Saltelli, K. Chan, E.M. Scott, Sensitivity Analysis, John Wiley Sons Publishers, New York, 2000. [69] J. Hirsch, DOE-2, Lawrence Berkeley National Laboratory, Camarillo, USA, 2012. [70] J. Hirsch, Manual DOE-2 Building Energy Use and Cost Analysis Program, Lawrence Berkley National Laboratory, Berkley, 2004. [71] UBA, Excel-Tool für das Projekt Energieeffizienz in Zahlen, Umweltbundesamt, Dessau-Roßlau, Germany, 2011. [72] BBSR, BBSR Wohnungsmarktprognose 2025, Bundesinstitut für Bau-, Stadtund Raumforschung, Bonn, Germany, 2011. [73] DESTATIS, Bauen und Wohnen, Mikrozensus – Zusatzerhebung 2006, Bestand und Struktur der Wohneinheiten, Wohnsituation der Haushalte, Fachserie 5/Heft 1, Statistisches Bundesamt, Wiesbaden Germany, 2008. [74] DESTATIS, Bauen und Wohnen Baugenehmigungen/Baufertigstellungen u. a. nach der Gebäudeart Lange Reihen z. T. ab 1960, Statistisches Bundesamt, Wiesbaden, Germany, 2010. [75] DESTATIS, Gebäude und Wohnungen – Bestand an Wohnungen und Wohngebäuden, Bauabgang von Wohnungen und Wohngebäuden, Lange Reihen ab 1969–2013, Statistisches Bundesamt, Wiesbaden, Germany, 2013. [76] Bayern LB, Deutschland bis 2040 Langfristige Trends und ihre Bedeutung für den Immobilienmarkt, Munich, Germany (2009). [77] M. Günther, L. Hübl, Wohnungsmangel in Deutschland? Regionalisierter Wohnungsbedarf bis zum Jahr 2025, Eduard Pestel Institut, Hannover, Germany, 2009. [78] S. Wanjani, A. Bauer, Bottom-up Characterization of the Residential German Building Stock – Reference Buildings and Energy Demand, Department of Energy and Environment, Chalmers University of Technology, Gothenburg, Sweden, 2012. [79] N. Diefenbach, R. Born, Basisdaten für Hochrechnungen mit der Deutschen Gebäudetypologie des IWU, Institut Wohnen und Umwelt, Darmstadt, Germany, 2007. [80] N. Diefenbach, H. Cischinsky, M. Rodenfels, K.-D. Clausnitzer, Datenbasis Gebäudebestand Datenerhebung zur energetischen Qualität und zu den Modernisierungstrends im deutschen Wohngebäudebestand, Institut Wohnen und Umwelt and Bremer Energie Institut, Darmstadt, Germany, 2010. [81] T. Loga, N. Diefenbach, R. Born, Deutsche Gebäudetypologie, Institut Wohnen und Umwelt, Darmstadt, Germany, 2011. [82] S. Klauß, W. Kirchhof, J. Gissel, Erfassung regionaltypischer Materialien im Gebäudebestand mit Bezug auf die Baualtersklasse und Ableitung typischer Bauteilaufbauten, Zentrum für Umweltbewusstes Bauen e.V., Kassel, Germany, 2009. [83] IWU, Deutsche Gebäudetypologie Systematik und Datensätze, Institut Wohnen und Umwelt, Darmstadt, Germany, 2005. [84] R. Born, N. Diefenbach, T. Loga, Energieeinsparung durch Verbesserung des Wärmeschutzes und Modernisierung der Heizungsanlage für 31 Musterhäuser der Gebäudetypologie, Institut Wohnen und Umwelt, Darmstadt, Germany, 2003.

A. Martinez Soto, M.F. Jentsch / Energy and Buildings 128 (2016) 38–55 [85] D. Walberg, A. Holz, T. Gniechwitz, T. Schulze, Wohnungsbau in Deutschland – 2011 Modernisierung oder Bestandsersatz, Arbeitsgemeinschaft für zeitgemäßes Bauen e.V., Kiel, Germany, 2011. [86] DESTATIS, Bevölkerung und Erwerbstätigkeit – Bevölkerungsfortschreibung, Fachserie 1, Reihe 1.3, Statistisches Bundesamt, Wiesbaden, Germany, 2012. [87] M. Demary, M. Voigtländer, Inmobilien 2025. Auswirkungen des demographischen Wandels auf die Wohn- und Büroimmobilienmärkte, Institut der deutschen Wirtschaft Köln (IW), Köln, Germany, 2009. [88] DIN, Energetische Bewertung von Gebäuden – Berechnung des Nutz-, Endund Primärenergiebedarfs für Heizung, Kühlung, Lüftung, Trinkwarmwasser und Beleuchtung – Teil 10: Nutzungsrandbedingungen, Klimadaten, DIN V 18599-10:2011-12, DIN Deutsches Institut für Normung e. V., Berlin, Germany, 2011. [89] DWD, Climate Data Center, Deutscher Wetterdienst Klima und Umwelt, Offenbach, Germany, 2015. [90] G. Hauser, G. Hausladen, Energiekennzahl zur Beschreibung des Heizenergieverbrauchs von Gebäuden, Baucom Verlag, Böhl-Iggelheim, Germany, 1990. [91] G. Hauser, G. Hausladen, Energiepaß Energetische Bewertung von Wohngebäuden mit Hilfe einer Energiekennzahl, Abschlußbericht, Fraunhofer IRB Verlag, Stuttgart, Germany, 1993. [92] G. Hauser, G. Hausladen, F. Otto, Energieeinsparung im Gebäudebestand, Baucom Verlag, Böhl-Iggelheim, Germany, 1996. [93] T. Loga, U. Imkeller-Benjes, Energiepaß Heizung/Warmwasser Energetische Qualität von Baukörper und Heizungssystem, Institut Wohnen und Umwelt GmbH, Darmstadt, Germany, 1997. [94] L. Joos, Energieeinsparung in Gebäuden, Vulkan-Verlag GmbH, Essen, Germany, 2004. [95] T. Loga, N. Diefenbach, J. Knissel, R. Born, Kurzverfahren Energieprofil, Institut Wohnen und Umwelt GmbH, Darmstadt, Germany, 2005. [96] P. Tzscheutschler, M. Nickel, I. Wernicke, H.G. Buttermann, Energieverbrauch in Deutschland – Stand 2007: Daten, Fakten, Kommentare, BWK – Das Energie-Fachmagazin 61 (6) (2009) 6–14. [97] Prognos AG, Öko-Institut, Modell Deutschland – Klimaschutz bis 2050: Vom Ziel her denken, Prognos AG/Öko-Institut, Basel, Switzerland/Berlin, Germany, 2009. [98] M. Beer, R. Corradini, C. Fieger, T. Gobmaier, L. Köll, R. Podhajsky, M. Steck, M. Zotz, Endbericht Energiezukunft 2050, Forschungsstelle für Energiewirtschaft e.V., München, Germany, 2009. [99] M. Beer, R. Corradini, T. Gobmaier, L. Köll, R. Podhajsky, G. Vogler, M. Zotz, CO2 -Verminderung in Deutschland, Forschungsstelle für Energiewirtschaft e.V., Munich, Germany, 2009. [100] W. Stiftung, Energieeffiziente Architektur, Krämer Verlag, Stuttgart, Germany, 2010. [101] UBA, Prozessorientierte Basisdaten für Umweltmanagement-Instrumente, Umweltbundesamt, Dessau-Roßlau, Germany, 2013. [102] M. Bost, B. Meyer, B. Hirschl, Überarbeitung der Vergabegrundlage für das Umweltzeichen “Warmwasserspeicher (RAL-UZ 124), Institut für ökologische Wirtschaftsforschung, Berlin & Institut für Ökologie und Politik GmbH, Hamburg, Germany, 2011. [103] DESTATIS, Bautätigkeit und Wohnungen Mikrozensus – Zusatzerhebung 2006 Bestand und Struktur der Wohneinheiten Wohnsituation der Haushalte, Statistisches Bundesamt, Wiesbaden, Germany, 2008.

55

[104] Die Entwicklung der Energiemärkte bis zum Jahr 2030: Energiewirtschaftliche Referenzprognose −Kurzfassung, Energiereport IV, in: W. Schulz (Ed.), Energiewirtschaftliches Institut an der Universität zu Köln, Cologne, Germany & Prognos AG, Basel, Switzerland, 2005. [105] M. Frondel, P. Grösche, H. Tauchmann, B. Krause, A. Wolf, D. Pantigoso, Erhebung des Energieverbrauchs der privaten Haushalte für das Jahr 2003, Rheinisch-Westfälisches Institut für Wirtschaftsforschung (RWI) Essen & forsa Gesellschaft für Sozialforschung und statistische Analysen mbH, 2005. [106] DESTATIS, Energieverbrauch der privaten Haushalte – Wohnen, Mobilität, Konsum und Umwelt, Statistisches Bundesamt, Berlin, Germany, 2008. [107] W. Ebel, W. Eicke-Hennig, W. Feist, Energieeinsparung bei Alt und Neubauten, Müller Verlag, Heidelberg, Germany, 2000. [108] K.-D. Clausnitzer, J. Gabriel, N. Diefenbach, T. Loga, W. Wosniok, Ermittlung von Effekten des KfW-CO2 -Gebäudesanierungsprogramms – Entwicklung der Methodik und Ergebnisse der Berichtsperioden 2005 und 2006, Bremer Energie Institut, Institut Wohnen und Umwelt and Institut für Statistik der Universität Bremen, Bremen, Germany, 2007. [109] UBA, Energieeffizienzdaten für den Klimaschutz, Umweltbundesamt, Dessau-Roßlau, Germany, 2012. [110] EnEV, Verordnung über energiesparenden Wärmeschutz und energiesparende Anlagentechnik bei Gebäuden (Energieeinsparverordnung – EnEV) vom 24.07.2007, BGBl. I S. 1519, zuletzt geändert 24.10.2015, BGL1 I. S. 1789. [111] IGCS, Long-Term Energy and Development Pathways For India, Indo-German Centre for Sustainability, Chennai, India, 2014. [112] G. Bazán Navarrete, E. Nava Palma, Anexo 14: Prospectiva energética para el periodo 2014–2030 del sector energía de, México Secretaría de Energía, 2005. [113] CEPLAN, Proyecciones de la Matriz Energética al Largo Plazo, Centro Nacional de Planeamiento Estratégico, Lima, Perú, 2011. [114] B. Limmeechokchai, S. Chawana, Sustainable energy development strategies in the rural Thailand: the case of the improved cooking stove and the small biogas digester, Renew. Sustain. Energy Rev. 11 (5) (2007) 818–837. [115] J. Lin, B. Cao, S. Cui, W. Wang, X. Bai, Evaluating the effectiveness of urban energy conservation and GHG mitigation measures: the case of Xiamen city, China, Energy Policy 38 (9) (2010) 5123–5132. [116] B. Chateau, B. Lapillonne, M. Alinhac, B. Laponche, G. Moury, La demande d’énergie finale de la France á l’horizon 2000: trois scénarios par la méthode MEDEE, Institut Economique et Juridique de l’Energie, Paris, France, 1979. [117] DESTATIS, 13. koordinierte Bevölkerungsvorausberechnung für Deutschland, Statistisches Bundesamt, Wiesbaden, Germany, 2015. [118] WärmeschutzV, Verordnung über den energiesparenden Wärmeschutz bei Gebäuden (Wärmeschutzverordnung – WärmeschutzV) vom 16.08.1994, BGBl. I S. 2121. [119] DCLG, The English House Condition Survey 2001, Department of Communities and Local Government, London, UK, 2001. [120] DCLG, The English House Condition Survey 2007, Department of Communities and Local Government, London, UK, 2009. [121] DCLG, English Housing Survey (EHS), Department of Communities and Local Government, London, UK, 2011. [122] S. Natarajan, J. Padget, L. Elliott, Modelling UK domestic energy and carbon emissions: an agent-based approach, Energy Build. 43 (10) (2011) 2602–2612.