Energy demand profile generation with detailed time resolution at an urban district scale: A reference building approach and case study

Applied Energy 193 (2017) 243–262

Contents lists available at ScienceDirect

Applied Energy journal homepage: www.elsevier.com/locate/apenergy

Energy demand profile generation with detailed time resolution at an urban district scale: A reference building approach and case study Georgios Kazas 1, Enrico Fabrizio ⇑, Marco Perino Politecnico di Torino, DENERG, Corso Duca degli Abruzzi 24, Torino 10129, Italy

h i g h l i g h t s
A bottom-up engineering model (DiDeProM) for thermal energy demand profiles generation is developed.
DiDeProM relies on samples of representative buildings technique.
Parametric analysis at building and district scale is carried out.
A stochastic aggregation method to generate a district thermal energy demand profile is applied.
Hourly time step for thermal storage technologies and demand side management studies can be obtained.

a r t i c l e

i n f o

Article history: Received 8 July 2016 Received in revised form 28 January 2017 Accepted 30 January 2017

Keywords: Energy demand profiles Building stock modelling District scale Hourly thermal energy demand profiles Demand side management

a b s t r a c t The energy demand in urban areas has increased dramatically over the last few decades because of the intensive urbanization that has taken place. Because of this, the European Union has introduced directives pertaining to the energy performance of buildings and has identified demand side management as a significant tool for the optimization of the energy demand. Demand side management, together with thermal energy storage and renewable energy technologies, have mainly been studied so far at a building scale. In order to study and define potential demand side management strategies at an urban scale, an integrated urban scale assessment needs to be conducted. DiDeProM, a model that can be used to generate detailed thermal energy demand profiles, at an urban district scale, has been developed in the current study. It is a bottom-up engineering model, based on samples of the representative building technique. A parametric analysis of the important variables of building energy performance at an urban scale has then been carried out. This has generated a database of normalized thermal energy demand profiles with an hourly time resolution. The final step of the process includes the generation of a detailed overall thermal energy demand profile at an urban district scale. DiDeProM was applied to a block of buildings in Turin (Italy) as a case study. After the calibration of the simulation model on real monitored data, a parametric analysis on 300 scenarios for a reference building was conducted, generating a database of seasonal thermal heating energy demand profiles with hourly time steps. An average hourly heating profile was generated from this database according to a specific aggregation approach. The DiDeProM application indicated that the model works properly at the scale of a typical small block of buildings, and it is able to generate a total thermal energy demand profile, with detailed time resolution, at an urban district scale. These profiles will be used to create demand side management strategies that will integrate thermal energy storage and r‘enewable energy technologies at a district scale. Ó 2017 Elsevier Ltd. All rights reserved.

1. Introduction Approximately half of the earth’s population lives in urban areas, and this percentage is going to increase due to a rise in ⇑ Corresponding author. 1

E-mail address: [email protected] (E. Fabrizio). Tessaloniki (Greece) 30 July 1985, Torino (Italy) 16 July 2016.

http://dx.doi.org/10.1016/j.apenergy.2017.01.095 0306-2619/Ó 2017 Elsevier Ltd. All rights reserved.

population in developing countries, such as Brazil, and India [1]. The main consumer of energy in urban areas is the building sector. In the European Union (EU), the building sector is responsible for 40% of the total energy consumption [2]. The European Union has been developing various actions to reduce the on-going energy demand and consumption increase, as well as to comply with the goals of the Kyoto Protocol. One of

244

G. Kazas et al. / Applied Energy 193 (2017) 243–262

Nomenclature

P ðRB;n ;Sk;g;m Þ

conditioned floor area of a building, m2 conditioned floor area of RB;n , m2 building n ! cluster of p Sk;g;m in the database typical monthly day database of normalized profiles sub-database of normalized profiles for each RB;n urban district normalized generated profile with an hourly time step, W m2 profile with an hourly time step related to Sk;g;m , W

P Bx

generated profile of a building, W

PI

profile of the urban district, W

P I;av g RB;n Ri

average profile of the urban district, W reference building daily mean solar radiation, W h m2

ABx ;cond: ARB;n ;cond: Bn C n;Sk dtyp D Dn I ! p ðRB;n ;Sk;g;m Þ ! ! ! !

these actions is the Directive on the Energy Performance of Buildings (EPBD), which came into force in 2002 (Directive 2002/91/EC) [3]. The goals of this Directive are to reduce energy consumption and reduce greenhouse gas emissions (GHGs) in the building sector, as well as to increase the share of the energy production by means of renewable energy technologies (RETs) by the year 2020. Therefore, the EU has developed actions, such as the creation of a common general framework for the assessment of the energy performance of buildings, the application of the minimum energy performance requirements to new buildings, the increase in the number of nearly zero-energy buildings (nZEB), the introduction of HVAC system inspections of buildings and energy certification of the buildings [3]. The tools that were used, or are at present under research and development for these purposes, include the building occupant and user awareness to energy efficiency concept, the refurbishment of existing buildings with low energy techniques, the construction of new low energy buildings, the optimization of energy systems with the integration of renewable energy and storage technologies, and the use of monitoring controls. However, in order to study the benefits of all of the aforementioned strategies, building models are necessary. The modelling and the prediction of the impact of efficient measures through simulations [4] are necessary because it is extremely difficult, or even impossible, to create an entire building or a building district in a laboratory with the purpose of conducting tests. The modelling and the development of tools to assess and improve the efficiency of individual buildings in urban areas has been underlined as a priority and an important challenge for the European Union’s environmental policy in the 21st century. The building models that have so far been developed for this purpose can be classified as physical, statistical or hybrid [5]. Physical models are very detailed, physics-based building models that investigate the energy performance of a building in terms of natural ventilation, heating, cooling etc. These models include the CFD approach, the zonal and the multi-zone or nodal approach. Statistical models are models that were developed using statistical approaches and methods. These models do not need the physical details of a building as input and do not need any physics or heat transfer equations of the buildings. Statistical models are based on a collection of large databases of measured quantities (e.g. energy consumption, econometric values and meteorological data), and include conditional demand analysis CDA, genetic algorithms, artificial neural network, etc. Hybrid models are models that couple statistical and physical models. These three types of models

!

Rm Ss;k Sv ;g Sl;m Sk;g;m SN DT i SN DRi Ti

average monthly solar radiation, W h m2 shading scenario infiltration scenario internal load scenario parametric analysis scenarios serial number of the temperature variation serial number of the solar radiation variation daily mean temperature, °C

Tm TBM

average monthly temperature, °C thermal building model

!

Greek letters h fixed model parameters of the TBM DT i variation in temperature, °C DRi variation in solar radiation, W h m2

are currently well developed at an individual building scale, and are used extensively to assess energy retrofitting measures, to predict future energy consumption, to mitigate carbon emissions and to develop new technologies. However, the interest of the EU over the last few years has not only focused on the building scale, but also on the urban scale, in order to create models for an integrated city-scale energy performance assessment [6–8]. This is because an investigation at a building scale does not represent a reliable approach to the behaviour of a building at an urban scale, where the interactions between the buildings in a neighbourhood or between urban districts represent a crucial parameter in the assessment of building energy behaviour. One of the main aims of the CI-NERGY European Project [9], which studies and develops methodologies and tools that can be used for integrated energy management at an urban scale, is to develop a strategy for the integration of demand side management with thermal energy storage technologies at an urban district scale. In order to study the integration of thermal energy storage and demand side management techniques, it is necessary to generate thermal energy demand profiles, with a detailed time step, at an urban district scale. Within the same project, a characterization of domestic hot water end-uses for integrated urban thermal energy assessment and optimization was developed [10]. A methodology for the generation of an overall thermal energy demand profile, with a detailed time resolution at an urban district scale, has been developed in this paper. A review of building stock modelling approaches has been carried out in the next section. Top-down and bottom-up modelling approaches are presented in the review. The review has mainly been focused on bottom-up modelling approaches, and the different techniques that are used to create a bottom-up engineering model. After the review, a bottom-up physics-based building model is developed, presented and applied to a case study in order to generate an overall thermal energy demand profile, with an hourly time-step, at an urban district scale. A data post-processing has been developed and presented. This post-processing is in fact a methodology for the identification and selection of typical monthly days. The postprocessing is a very important step in the overall methodology as it is used to obtain a synthetic representation of detailed thermal energy demand profiles for demand side management and active thermal energy storage integration at an urban district scale [11,12].

G. Kazas et al. / Applied Energy 193 (2017) 243–262

2. Review of building stock modelling approaches There are many reasons why carrying out an energy assessment at a higher scale than the building one, is important to investigate technologies, strategies and policies at the building stock scale. A large number of building stock models have been developed over the last few years. These models, including physical, data driven and hybrid models, mainly dealt with the residential sector of urban areas. However, building stock models were also used for some non-domestic, commercial and public purposes. The largest amount of published literature is on residential building stock models, due to the important role of the residential sector in the total energy consumption in urban areas (in the European Union, 63% of the final energy consumption of the total building stock comes from the residential building sector [13]). Another reason for the increased interest in residential building stock models concerns the difficulties and the impossibility of determining the uses and the behaviour of the occupants of the large and diverse nondomestic sector [14]. Table 1 presents the top-down and bottom-up building stock modelling approaches [15] and a further possible breakdown of the approaches. These building stock modelling approaches cover two different study levels. The top-down modelling approach has been developed to study the building stock at an aggregated level, such as a national building stock, and includes mainly data-driven models. On the other hand, the bottom-up modelling approach has been developed to assess the energy performance of a building stock, starting from a disaggregated level, such as end-use modelling, and it includes statistical and engineering (physics-based building) approaches and methodologies. 2.1. Top-down modelling approach The top-down modelling approach studies the building stock as a whole at an aggregated level. It does not investigate individual building models, technologies or energy end-uses. Top-down models are in fact data-driven models, and were developed in order to investigate the energy consumption of buildings and to associate it with historical, economic and technological long-term changes in the building sector. They mainly use easily found aggregated data, such as gross domestic product (GDP), or other econometric variables, climate parameters, construction parameters and estimations of the number and the use of the appliances [16]. According to the kind of input data they use, top-down modelling approaches can be divided into two sub-categories: econometric and technological models. Econometric models assess the energy consumption related to economic and macroeconomic parameters, such as fuel prices, income, gross domestic product etc. The main drawback of econometric models is that they only focus on historical, economic and statistical data, and do not take into account physical and technological factors or their evolutions. When there is a lack of information on the evolution of the technological factors, it is difficult to reliably predict the energy demand, energy consumption and CO2 emissions of the building stock, because the future environmental, technological and economic conditions may be different from those of the past.

Table 1 Building stock modelling approaches. Top-down Bottom-up

Econometric Technological Statistical Engineering (physics-based building)

Data driven

Physical

245

Technological models assess the energy performance of a building stock using technological data, such as technological progress, as well as architectural and structural changes. These models investigate the technological parameters at an aggregated level, and mainly consider the national building stock [17]. It is difficult to precisely predict the energy demand, because the technological top-down model parametrization is not carried out at a detailed level. An example of a typical technological non-residential top-down model can be found in [18]. The authors developed a top-down model to benchmark the energy performance of the school sector in the UK. They in particular gathered data on the building characteristics of schools and created an aggregated database of the school sector. In order to predict the consumption of the building stock, they developed a statistical model, using the Artificial Neural Networks (ANN) technique. The results of the top-down model were compared with those of other bottom-up models of the same non-domestic building stock. The most significant outcome of that paper was that the top-down model was less precise than the bottom-up one. The level of preciseness depends on the amount of data that is available. As a result, the authors proposed the combined use of top-down and bottom-up models to increase the level of preciseness. As mentioned above, the top-down modelling approach is a method that can be used to assess the overall energy consumption of a building stock, and to compare it with economic and other statistical factors. The top-down modelling approach can also be used to determine the supply requirements [19]. However, this technique suffers from some limitations and disadvantages, in comparison to the bottom-up approach, as it is not possible to conduct a sub-system modelling or assessment, and a huge amount of data is needed. These limitations make the top-down modelling approach less accurate for technical purposes, such as the prediction of the energy demand and the investigation of demand side management strategies. Furthermore, this approach is aimed at assessing the overall seasonal or annual energy demand, and does not usually provide information about the time profiles of the energy demand.

2.2. Bottom-up modelling approach The bottom-up modelling approach investigates the building stock at a detailed level. Bottom-up models are models in which the energy performance investigation of a building stock is started at a disaggregated level, such as a building sub-system level. Later, by resorting to various statistical or deterministic aggregation methods and techniques, the energy values are aggregated to represent the entire building stock. The level of aggregation can either be at an urban or a national scale. The bottom-up modelling approach uses data from a hierarchal level, and not only from the whole building sector, like the top-down approach does. Energy demand prediction, demand side management strategy development, climate change and environmental issue evaluation and mitigation can be carried out with the bottom-up modelling approach at various scales; from the single end-user, individual building and block of buildings scale to the urban district, regional or national building stock scale [16]. In general, the bottom-up modelling approach makes use of very detailed models. The level of detail depends on the level of disaggregation. This approach includes data-driven, physical and hybrid models, and it needs extensive empirical data databases, due to the required level of detail, in order to develop and validate the models themselves and to obtain reliable results [17]. Thus, bottom-up models may be classified into two sub-categories.

246

G. Kazas et al. / Applied Energy 193 (2017) 243–262


Bottom-up statistical models
Bottom-up (physics-based) engineering models Statistical bottom-up models are data-driven models that are based on historic and energy performance data to establish the relationship between end-uses and energy consumption. These models can estimate the energy consumption of a representative cluster of buildings from the neighbourhood scale to the city and national scale. Statistical models assess the energy performance and predict the energy consumption, using statistical techniques [20], such as regression [14,21–25]. In general, the bottom-up statistical modelling approach does not provide detailed information on the end-uses of buildings, such as the energy consumption, the use of appliances, or heating and cooling applications [19]. The statistical modelling approach needs a huge amount of data, which can be obtained from either public or private databases, in order to be reliable. Engineering or bottom-up physics-based building models are physical models in which the energy performance assessment is carried out using heat transfer and thermodynamic equations, considering the use of appliances and equipment as well as the power rating. These models are very detailed and are able to evaluate the energy performance of each end-user [16]. The bottom-up engineering modelling approach is basically developed through clustering and classification techniques, including archetypes [26] or samples of representative buildings. The bottom-up engineering modelling approach is more suitable than the top-down approach for a detailed study and prediction of energy demand profiles in order to investigate energy technologies, such as thermal energy storage. For this reason, a review on the existing bottom-up engineering models has been carried out in the next section. A physics-based bottom-up engineering methodology has then been proposed, developed and applied to a case study.

2.2.1. Bottom-up engineering modelling approach In order to predict the energy demand, to assess the energy performance of buildings and districts, and to apply energy demand and CO2 emission reduction strategies and policies, it is necessary to develop a bottom-up engineering modelling approach. The bottom-up engineering modelling approach consists of physical (white box) models. These models include physical multi-zone or more detailed models of buildings that are considered representative of a class of buildings. These buildings are simulated be means of dynamic simulation software, such as EnergyPlus, TRNSYS, ESP-r, and IDAICE, in order to predict energy demand values, temperatures or other energy related values. Then, according to the aggregation method adopted in each model, the data are transferred to a higher level in order to be representative of a neighbourhood, block of buildings, city or national building stock. The input data that are required for this method are physically measurable data such as: geometrical and architectural elements (façade, walls, roofs and windows) with their thermal properties (U value), the configuration and characteristics of the energy systems, infiltration and ventilation rates, heating and cooling setpoints, internal-external temperatures, climatic conditions, occupants’ behaviour and ownership and the use of the appliances, etc. [17–27]. A physics-based bottom-up engineering model can be developed through the use of three techniques:
‘‘Brute-Force”
Building Archetypes
Samples of representative buildings

The first technique involves the modelling of each individual building, while the other two involve the classification of the building stock into clusters, using cluster analysis techniques, and then the development of representative physical building models. ‘‘Brute Force” The term ‘‘Brute Force” is used in this study to describe the first set of models. It is borrowed from computer science, and is mainly used in optimization studies [4–30]. In this paper, the ‘‘Brute Force” technique is defined as a technique that can be used to develop bottom up engineering models, which includes the modelling and simulation of each single building of a building stock. The simulation is carried out, by means of dynamic energy simulation software, at various scales, such as a block of buildings or a neighbourhood scale. With this method, the energy performance of each single building is aggregated in order to achieve the overall energy performance of a particular building stock. The most frequently used aggregation method is the one that considers the sum of the energy values of all the buildings in the building stock. This is a completely deterministic and detailed method and it is, at least theoretically, highly accurate. Each single building in the stock is considered an individual physical building model (whitebox). The main disadvantage of this method emerges when the research scale becomes wider than a small block of buildings or a small neighbourhood. When the building stock is relatively large, the use of the ‘‘Brute Force” technique becomes time consuming. The ‘‘Brute Force” technique is in fact a very computational intensive method for the assessment of the energy performance of the house stock of an urban district, city or a country, as it requires a huge amount of resources. However, the ‘‘Brute Force” technique can be considered a suitable method for the investigation of small building stocks, such as small neighbourhoods [31,32] and very small blocks of buildings. An energy assessment of a block of buildings belonging to the Royal Botanic Gardens at Kew Gardens in the UK was carried out in [33] in order to evaluate retrofitting measures and to calculate the energy consumption reduction and CO2 emission mitigation. Each of the three Kew Garden buildings (Palm House, Temperature House and the Princess of Wales Conservatory) was modelled and simulated individually in order optimize their district energy supply. It is easy to understand that in order to investigate a larger building stock than just a few buildings, it is necessary to classify the buildings of the stock into classes and define archetypes of buildings. 2.2.1.1. Building archetypes. In the literature, the most common practice for an energy performance assessment of building stocks is the identification, development and use of building archetypes. This method includes the classification of the building stock into clusters of buildings [20]. According to predefined common variables, such as geometrical and architectural parameters, age, use etc. Building archetype models are developed from the clusters as statistical combinations of the average parameters of all the buildings belonging to the same class. A representative example of an archetypes development methodology is presented in [34]. The archetypes are then simulated, prevalently in multi-zone energy simulation software tools, in order to calculate the total and seasonal energy consumption and temperatures, and to establish and evaluate retrofitting actions, carbon mitigation scenarios and policies. These individual archetype predicted energy values are then aggregated by multiplying them by the total number of the buildings within the same cluster in order to obtain the overall energy value of the building stock. This technique has mainly been used in energy decision making and policy tools. The seasonal or annual energy consumption and

G. Kazas et al. / Applied Energy 193 (2017) 243–262

CO2 emissions were calculated in order to apply retrofitting and climate change mitigation measures. Some notable examples of bottom-up engineering models in which the building archetypes technique was used, and which were developed for energy policy tools, are the EEEP model [35], DECM [15], a model for the Scottish building stock [36], ECABS [37–39], REMA [40] and the model developed by Wang and Holmberg [41] for the retrofitting of the Swedish residential building stock. The aforementioned models could also be classified as hybrid bottom-up engineering models, because they are a combination of physical models and data driven models. Back in the 1990s, a bottom-up model was developed by Shorrock and Dunster [42] in order to assess the energy use of the building stock in the UK. This model was called BREHOMES and was developed to predict future energy consumption and carbon dioxide emissions. BREDEM simulation software was used to calculate the energy consumption of the archetypes. The aggregation is carried out by multiplying the energy consumption of the archetypes by the number of the buildings in the same cluster. Huang and Brodrick [43] created another model for the US building stock to calculate the total heating and cooling consumption. This model was created in order to apply energy efficiency measures. Other similar bottom-up archetype models are the Belgium stock model developed in [44], and the Finnish building stock model developed in [36,37], which were used to investigate the entire national building stocks. The engineering modelling approach based on archetype technique was developed to assess large building stocks, for example, entire national building stocks. They are a combination of physical models and data driven models that were created to apply retrofitting scenarios. These models deal above all with the prediction of the total annual energy consumption in order to investigate energy reduction measures and policies. 2.2.1.2. Samples of representative buildings. In this case, some buildings are selected as representative of a particular building stock and are modelled. It is possible to confuse these models with building archetypes, but there is a basic structural difference. This difference lies in the parameters considered in the building model. Samples of representative models are developed considering actual data. Therefore, these models can also be called ‘‘real building models”. Archetypes are those building models that are created as average statistical models of a cluster, instead of as samples of representative buildings that are the actual building models of existing buildings, which are selected from the building stock as being representative. The samples are prevalently used for huge building stocks with a large variety of building types, such as cities and national building stocks. It is necessary to use a large number of samples of representative buildings to have a reliable representation of the building stock. This is due to the high level of detail with which these models are constructed. When a model is very detailed, it represents a very small number of buildings that have the same parameters. For example, it is difficult to define the various occupants’ behaviour variables with just one representative building, even when the buildings belong to the same typology. On the other hand, one building archetype can represent a larger cluster of buildings, because it is a simplified building model, and its average parameters allow a large number of buildings to be represented. Therefore, this technique requires a large database of samples of representative building models. The representative buildings (real models) are simulated individually, and the results obtained by those building models are later aggregated. The number of the buildings in the buildings class multiplies the individual energy values. Characteristic examples of bottom-up engineering models that were developed with samples of representative buildings can be found in [45–48]. These models were developed for the energy

247

assessment of Osaka’s residential building stock. The authors used a large number of very detailed representative building models (460 types of dwellings) to simulate and calculate the energy consumption and CO2 emissions, and to evaluate some efficiency measures. Other bottom-up engineering models that use this technique are described in [44,49]. 2.3. Critical discussion The literature review on building stock modelling approaches has revealed weaknesses and gaps pertaining to the aggregation methods and the final goal of the investigation. As mentioned above, the aggregation methods that were used were basically simple methods that included sums and multiplications. Although the ‘‘Brute Force” technique can be considered a reliable but complicated method to predict the energy demand of the building stock, the other two methods resort to aggregation methods that reduce their preciseness. In these aggregation methods, the buildings in a class are assumed that have the same energy values as the archetype. Although this is a necessary assumption, there is an important difference in the other parameters that affect the energy performance of the buildings in a class that have the same typology. These parameters are variables, such as the internal loads and infiltration rates [50], that depend, for instance, on the occupants’ behaviour. The occupants’ behaviour varies not only from one building to another of the same class, but also between the different floors of the same building, or even between the different flats on the same floor. This variation is due to the fact that the occupants’ behaviour is influenced by various factors, such as environmental factors, building and system related factors, the occupants’ ages, gender, culture, education level and other factors like energy use awareness, price and time [51]. Many studies can be found in the current literature about the prediction and influence of occupants’ behaviour on building energy performance [41,42]. The great challenge as far as the occupants’ behaviour field is concerned, is its unpredictable nature. This is why much of the literature focuses on the development of prediction models of the occupants’ behaviour in residential buildings. These studies focus on statistical [52] and stochastic models [53] and on predicting how occupants act. However, the literature above all investigates the effects of occupants’ behaviour at a building scale. The state of the art review also underlined that the current models are used to determine the total energy consumption, annual and seasonal, and CO2 emission values. This is because the most frequent scope of these studies is to develop strategies for efficiency measures, as well as for retrofitting scenarios and policies. On the other hand, energy demand profiles, with detailed time steps, are needed to study energy systems and technologies and to plan efficient detailed strategies. In order to conclude the critical review of the literature, it should be pointed out that it is necessary to predict energy demand profiles of the building stock with a very detailed time resolution, taking into account the geographical and occupants’ behavioural diversity to study, model, size and assess energy systems and technologies, such as decentralized thermal energy storage technologies, at an urban scale. No detailed models that are able to predict thermal energy demand profiles, whether seasonal or annual, at an urban district scale, and which take into account all the aforementioned variables, were found in the literature. 3. Scope of the study The current study has focused on filling the gaps that were identified in the state of art review. An engineering bottom-up

248

G. Kazas et al. / Applied Energy 193 (2017) 243–262

model has been developed. This model, which is called DiDeProM (District, Demand, Profiles Model), is based on the sample of representative building approach. The scope of DiDeProM is to develop a model (methodology) that will be able to generate a common thermal energy Demand Profile at a district scale. DiDeProM consists of two steps. The first step is the parametric analysis of the significant variables that affect the energy performance of a building, such as the shading profiles, internal loads and infiltration rates. This parametric analysis was conducted on the considering detailed reference building (white-box) models. It is worth mentioning that, at this stage, the proposed methodology takes into account the occupants’ behaviour at a district scale, but with some assumptions, without applying occupants’ behavioural prediction models. The various building and thermal characteristics of a district are also taken into account in this parametric analysis. The parametric analysis leads to a database of thermal energy demand profiles that are generated from many scenarios of the samples of representative buildings. The second step consists of the application of a stochastic aggregation approach, in order to generate an average representative thermal energy demand profile at a district scale, with detailed time resolution. The database is used as a box in which profiles for the buildings of a district can be chosen randomly. After a statistical interpretation of the selected profiles, a stochastic aggregation method was applied to generate the average district profile. The outcome of this aggregation method is a representative average thermal energy demand profile, with an hourly time step, at an urban district scale. Such a profile represents a prerequisite and fundamental knowledge for an investigation, sizing and analysis of decentralized thermal energy storage, TES, and renewable energy technologies, RETs, at an urban district scale.

4. Methodology for the generation of thermal energy demand profiles at a district scale (DiDeProM) 4.1. DiDeProM overview The methodology consists of two steps. The first step (Fig. 1) is the creation of the thermal energy demand profile database, and the second step (Fig. 2) is the application of the aggregation method to generate the average thermal energy demand profile of the district.

After the definition and analysis of the urban district configuration, buildings are clustered according to their typology into homogenous subsets. Each building subset is assumed to be represented (as far as its thermal and energy behaviour is concerned) by just one reference building, RB;n . In this way, a few reference buildings can be selected and modelled as being representative of the entire district. A parametric analysis is then carried out on these reference buildings, and critical variables on building energy performance are investigated. As a result of the parametric analysis, a database of normalized thermal energy demand profiles is created, D, according to the conditioned floor area (Fig. 1). Each of the buildings in the district is then matched to one of the reference buildings, and to one of the possible scenarios related to the shading configuration (Fig. 2); the latter aspect is based on a semideterministic approach. The matching to the building usage profiles is instead a random process (as will be explained in Section 4.3). At the end of the procedure, all the buildings in the district will be associated to an energy demand profile from the database, D. Finally, by summing all of these ‘‘single building” energy demand profiles, an overall profile of the energy demand !

of the district P I is created. The procedure of associating a building usage profile of the database to a specific buildings is repeated several times ðj timesÞ in order to obtain a large number of random !

selections and a reliable average district demand profile P I;av g . 4.2. Step 1- Parametric analysis and database of the energy demand profile formulation 4.2.1. Reference building selection and modelling The methodology starts with the selection of an urban district ðIÞ. The urban district is characterized by the Latin letter I, which comes from the Latin word ‘‘insula”. Insula is the term that was used in ancient Rome to describe a block of buildings, and it was the main structural element of urban planning in ancient Rome. In practice, an urban district, I; can be a block of buildings, a number of blocks of buildings or a municipality in a city. The scale of the Insula is defined according to the level of research that is carried out. An Insula, I, consists of a specific and finite number of buildings; it can be described mathematically by means of a bounded set ðBn Þ:

I ¼ fB1 ; B2 ; B3 ; B4 ; . . . ; Bn g

Fig. 1. Methodology flow chart (step 1).

ð1Þ

249

G. Kazas et al. / Applied Energy 193 (2017) 243–262

Fig. 2. Methodology flow chart (step 2).

After the selection of the Insula, I, the buildings belonging to I are investigated, clustered [54] and classified according to various parameters, such as their age and typology [55,56]. On the basis of the classification that is carried out, one or more buildings are selected as Reference Buildings RB;n ðwhereRB;n ¼ fBi ; Bj ; . . . ; Bx g # I; n : the number of the RB Þ. Each reference building is assumed to be representative of an entire cluster. The reference buildingsðRB;n Þ are then modelled in dynamic multi-zone energy simulation software and tools using the actual data of the real building. This simulation is based on detailed white-box models, as is normal practice in the samples of representative buildings technique [16]. These models could also be called Real Building Models because the model input parameters are in fact those of the actual reference building parameters. Therefore, the reference building, RB;n , differs from other archetypal building models, which instead are statistical average representations of a building stock. The drawback of the limited stock representation of real building models and the need for a large database are cancelled out by the parametric analysis proposed in the current methodology, as it will be shown in Section 4.2.2). 4.2.2. Parametric analysis of the reference buildings In order to use the smallest possible number of buildings to represent the entire Insula I and to cover various parameters pertaining to the urban districts and to the unpredictable behaviour of the occupants, a parametric analysis needs to be carried out in the RB;n domain. This parametric analysis results in a significant number of seasonal thermal energy demand profiles with an hourly time step, !

P . These profiles are used to create a large database. First, a number of possible scenarios pertaining to some critical building energy performance parameters should be considered, at a building and a building stock scale: the shading profiles of the buildings, the internal loads (such as the occupant density, lighting and electrical equipment loads) and the air infiltration rates. Such parameters are very important to determine the thermal behaviour

of a building and to take into account the diversity of the different buildings in an Insula. The shading profile is influenced above all by the relative position of the building (orientation) and the surrounding buildings.2 The internal loads and the infiltration rates are instead influenced to a great extent by the occupant’s behaviour, and their temporal and spatial dimensions may be different (e.g. they can vary from building to building and/or at a sub-building scale, such as at the floor level or at the single apartment level). In practice, three families of scenarios can be considered:
Shading Scenarios (Ss;k Þ,
Air Infiltration Scenarios (Sv ;g Þ
Internal Load Scenarios (Sl;m Þ Each building in the Insula, I, can be subject to operating boundary conditions that can depend on any combination of the above listed scenarios. Therefore, all the possible combination of Ss;k , Sv ;g , and Sl;m should be taken into consideration. The result is a data set of Parametric Analysis Scenarios, Sk;g;m , which are functions of the Shading, Infiltration and Internal Load Scenarios. The elements of this data set, Sk;g;m , constitute all the possible boundary conditions for the energy simulations:

Sk;g;m ¼ f ðSs;k ; Sv ;g ; Sl;m Þ

ð2Þ

Once the full set of scenarios has been defined (see Fig. 3), a simulation of each reference building belonging to RB;n is performed for each Sk;g;m scenario, by means of the detailed multizone model (white-model). The outcome of the parametric analysis is a number of thermal !

energy demand profiles. These profiles, P ðRB;n ;Sk;g;m Þ , are functions of the parametric analysis scenarios and of the fixed parameters, h 2 Only in the case of movable solar shading devices - installed at the building scale could there be a significant influence of the users’ behaviour.

250

G. Kazas et al. / Applied Energy 193 (2017) 243–262

Fig. 3. Graphical sketch of the procedure used to assess the scenario database.

(the ‘‘standard” building parameters of a thermal model). It is worth mentioning that the fixed parameters are identified from the features of the specific reference building, RB;n , while the scenarios are picked from the Sk;g;m set. These profiles constitute a bounded set, whose number of elements is equal to the total number of the scenarios, Sk;g;m , times the number of the reference buildings RB;n

is a better known and more stable variable, at a building scale, than the internal load and the air infiltration rates. Therefore, it can be chosen in a quasi-deterministic way using GIS software, satellite maps or similar tools, like the procedure reported in [57]. Thus, each cluster ðC n;Ss;k Þ in Dn consists of normalized thermal energy demand profiles (Eq. (5)) with a specific shading scenario (Ss;k Þ:

!

n o9 8 > for k ¼ 1;1 6 g 6 b;1 6 m 6 c;C n;Ss;1 : p! ðRB;n ;S1;g;m Þ > > > > > > n o> > > > > > < for k ¼ 2;1 6 g 6 b;1 6 m 6 c;C n;Ss;2 : p! ðRB;n ;S2;g;m Þ > = Dn ¼ ; 8Dn 2 D ... > > > >   > > > > > for k ¼ a;1 6 g 6 b;1 6 m 6 c;C n;Ss;a : p! ðRB;n ;Sa;g;m Þ > > > > > : ;

P ðRB;n ;Sk;g;m Þ ¼ TBM½Sk;g;m ; hðRB;n Þ½W

ð3Þ

where





!

P ðRB;n ;Sk;g;m Þ : thermal demand profile of a certain RB;n and Sk;g;m RB;n : reference building Sk;g;m : parametric analysis scenarios hðRB;n Þ: fixed parameters of the Thermal Building Model (TBM) for the RB;n reference building

4.2.3. Set-up of the database of normalized demand energy profiles !

The thermal energy demand profiles, P ðRB;n ;Sk;g;m Þ , are normalized according to the conditioned floor area ðARB;n ;cond: Þ of the chosen reference building. The normalization process is necessary to make the profiles suitable for adoption for other buildings that are analogous, in configuration, to the reference building, but which have a different size (e.g. floor area). ! p ðRB;n ;Sk;g;m Þ

!
 P ðRB;n ;Sk;g;m Þ W ¼ ARB;n ;cond: m2

ð4Þ

where: !


p ðRB;n ;Sk;g;m Þ : normalized thermal energy demand profile !


P ðRB;n ;Sk;g;m Þ : thermal energy demand profile obtained from the parametric analysis
ARB;n ;cond: : conditioned floor area of the reference building After the normalization of the thermal energy demand profiles, the database is created. The database of Fig. 4 consists of sub-databases, Dn , of normalized thermal energy demand profiles, one for each RB;n reference building. The profiles in each sub-set Dn are then sorted into clusters ðCÞ. The Shading Scenario (Ss;k Þ variable is selected to define the clusters. This is due to the fact that, as already mentioned, the shading profile

ð5Þ where
D ¼ fD1 ; D2 ; . . . ; Dn g: overall database that is created from the parametric analysis (one sub-set of the database for each reference building) !


Dn ¼ fa  p ðRB;n ;Sk;g;m Þ g: sub-sets of the database; one sub-set for each reference building n
a: number of Shading Scenarios selected as the main variable of the cluster Finally, the database, D, is used to generate an overall average thermal energy demand profile, through the stochastic aggregation approach, which is described in the next section. 4.3. Step 2- Aggregation method and generation of the urban district thermal energy demand profiles In order to generate an average profile of the entire urban district, I, an aggregation method has to be developed which: (1) associates each actual building of the district to a reference building, RB;n (2) identifies the subset, Dn , related to the reference building (3) ‘‘picks” the appropriate normalized energy demand profiles, !

p ðRB;n ;Sk;g;m Þ , from the sub-database, Dn (this takes place in two stages: 3a - related to the Solar Shading Scenarios - and 3b related to the Internal Loads and Air Infiltration Scenarios) (4) sums (hour by hour) all the hourly profiles.

251

G. Kazas et al. / Applied Energy 193 (2017) 243–262

Although this could seem the best possible procedure, it can rarely be applied to real cases, since very detailed knowledge of the district is needed. Such knowledge is seldom available.

Fig. 4. Overall Database and sub-databases.

Once the actual building has been matched to the appropriate reference building, RB;n , and, consequently, the Dn sub-set has been identified (points 1 and 2), the shading scenario (Ss;k Þ and the conditioned floor area of each building ðAB;cond: Þ of the urban district can be defined. The Shading Scenario can easily be determined for newly constructed buildings from construction plans, while it can be assessed by means of satellite maps or GIS tools for already existing buildings. It is then possible to move to point 3, which is divided into two phases (i.e. 3a and 3b). First (stage 3a), it is possible to identify a specific cluster, C n;Ss;k , of normalized energy demand profiles within the Dn subset on the basis of the knowledge of the shading scenario. Thus, each actual building in the Insula is matched, according to its corresponding RB;n and shading scenario, to a specific cluster of normalized hourly energy demand profiles, that is characterized by a specific shading configuration. In the next phase (stage 3b), the influence of the internal loads and of the air infiltration rates has to be introduced. As previously mentioned, these quantities are influenced a great deal by the occupant’s behaviour, and they may be different in temporal or spatial dimension (e.g. they can vary from building to building and/or at a sub-building scale, such as at the floor level or at the single apartment level). It is difficult, if not impossible, to predict and control human behaviour. The extensive variety and the unpredictable nature of the occupants’ behaviour creates large differences and uncertainties in energy building simulations [15,58,59]. Consequently, in order to have representative profiles for various energy uses, a procedure needs to be established to identify appropriate internal loads and air infiltration scenarios in the C n;Ss;k cluster. Three different approaches could be followed for this purpose: I. deterministic II. stochastic, based on statistics of the population’s characteristics, such as a census III. fully stochastic.

4.3.2. The stochastic approach based on statistics of the population’s characteristics The population-statistics approach (II) could be used as a further step. If there is no detailed information on the family nuclei living in a district, it is possible to resort to statistical data referring to the overall population of the area, such as demographic statistics or a census [60]. In this way, information can be obtained, for example, on the percentage of families constituted by young couples without children, young couples with children, single people, retired people, etc. It is then possible to associate the pertinent internal load and infiltration rate scenarios to each category of inhabitants and to match these scenarios to those in the C n;Ss;k cluster, taking care to keep the same percentage of the profiles as that in the population statistic data (for example, if there is 30% of retired couples in the district, whose average behaviour is represented by the specific scenario Sa;b;c in cluster C n;Ss;a , then this scenario must be assigned to 30% of the representative buildings in the Insula, or – even better - to 30% of the heated volume of the Insula). This approach represents a trade-off between the (theoretical) better accuracy of the prediction and the level of detail that is needed to perform the analysis. Simplification and approximation are implied, since the statistics of the family typology do not necessarily correspond to the same statistics as those of the occupied heated volume or heated floor area (that is, 30% of the retired people could in reality occupy a built space that may represent just 20% of the overall heated volume of the Insula). 4.3.3. The fully stochastic approach In this case, no information about the users’ typology is required, and the whole procedure is based on a fully stochastic combination of internal loads and air infiltration scenarios. This is the approach that has been developed and used in the present paper to develop step 3b. It has the advantage of being applicable both during the planning of a district (when the users’ typology has not yet been defined) or when no information is available on the population living in the Insula. The methodology is based on the recursive selection of the energy demand profiles picked randomly from the C n;Ss;k cluster. !

In practice, a normalized energy demand profile p ðRB;n ;SS;g;m Þ is selected randomly from cluster C n;Ss;k (identified at step 3a) for each building in the Insula, which is represented by a specific reference building, RB;n . This normalized thermal energy demand profile is multiplied by the conditioned floor area of the real building ðABx ;cond: Þ to obtain one of the possible hourly time profiles of the energy demand of !

the building, P Bx : !

4.3.1. The deterministic approach If the aim of an investigation is to know the exact habits of each individual family living in a district, the internal load and the infiltration scenarios in the C n;Ss;k cluster could be chosen on the basis of an almost deterministic process, in which the profiles are chosen in order to match the behaviour of the occupants as much as possible.3 This is the so-called ‘‘deterministic approach”. 3

It is worth noting that it is probably necessary to apply a sort of weigh to the energy demand profiles, on the basis of the heated floor area or heated volume. In fact, it may happen that users with different behaviour are living in different flats in the same building, but the proposed procedure cannot be used below the whole building scale.

!

P Bx ¼ p ðRB;n ;Sk;g;m Þ  ABx ;cond: ½W

ð6Þ

The same calculation is repeated for all the other buildings in the Insula. After this first random selection, the thermal energy demand profiles are summed, and the overall thermal energy demand profile of the Insula is created: !

PI ¼

n ! X P Bx ½W

ð7Þ

x¼1

The procedure is shown graphically in Fig. 5. Clearly, the !

obtained energy demand profile, P I , represents just one particular case of the many possible ones.

252

G. Kazas et al. / Applied Energy 193 (2017) 243–262

Assuming that the occupants’ categories (each one corresponds to a certain combination of the Internal loads and the air infiltration profiles) are distributed according to a normal distribution over the Internal load and air infiltration scenarios, if this process is repeated several times, and the various specific energy demand !

profiles, P I;j , are averaged, it is possible to obtain an average thermal energy demand profile, with an hourly time step, which is statistically representative of the entire energy demand of the district. Mathematically, this translates into:

8 9 n ! X ! > > > > P ¼ P > I;1 Bx > > > > > > > 1 > > > > > > > > n ! X ! > > < = P I;2 ¼ P Bx ½W 1 > > > > ... > > > > > > n ! > > X ! > > > > > > P ¼ P B > > I;j x > > : ;

ð8Þ

!

where j is the number of selected random profiles !

The average thermal energy demand profile, ðP I;av g Þ, is then assessed from the set of the thermal energy demand profiles (8): !

!

!

!

4.4. Definition and selection of typical monthly days Finally, a post-processing of the representative district thermal

1

P I;av g ¼

demand profile, it is also possible to consider assuming the energy demand profile that represents the worst case (e.g. the one for which the demand is exceeded in less than 5% of the cases) and the best one (e.g. the statistically lowest energy demand profile of the district, which corresponds to about 95% of energy conscious occupants). The important variables that affect the energy performance and the unpredictable nature of these variables, at a district scale, are taken into account in this kind of demand profile generation. Although the analysis is carried out at an aggregated level (e.g. only a few representative reference buildings), its scale is rather detailed, because the different scenarios are defined with a small ‘‘granularity” of information (floor level).

!

P I;1 þ P I;2 þ P I;3 þ . . . þ P I;j ½W j

ð9Þ

!

Finally, P I;av g can be assumed to be a representative thermal energy demand profile, with an hourly time resolution, at a district scale. However, sensitivity studies can also be conducted by exploit!

ing all the information provided by the P I;J database. For example, !

apart from the average energy demand profile, P I;av g , it is also possible to adopt those profiles that correspond to the ±2r variances of the distribution of the energy demand profile distributions and to use these profiles to perform a ‘‘what-if” analysis. In other words, apart from the studies conducted with the average energy

energy demand profile P I;av g is required. The average thermal energy demand profile, with an hourly time step, which is generated by DiDeProM is a seasonal/annual thermal energy demand profile with an hourly time step. This profile is fundamental for developing calculations and/or simulations, but it is very difficult to represent it in a graphical way or to use it to perform statistical analyses, because of the detailed data that it contains. In order to investigate and quickly analyse energy technologies, such as thermal energy storage (short-term and long-term) and renewable energy technologies, and to apply energy demand strategies to obtain an efficient district energy performance, or to conduct sensitivity analyses, a concise and easy-to-use representation of the profile is sometimes necessary. Thus, a data postprocessing method was created in order to achieve a synthetic representation of the detailed thermal energy demand profiles. The proposed procedure adopts the ‘‘typical monthly day” concept. In the current project, a typical monthly day is defined as the day of the month that is closest to the average monthly temperature (T m ) and the average monthly rate of solar radiation ðRm Þ.

Fig. 5. Generation of thermal demand profiles at an urban district scale.

253

G. Kazas et al. / Applied Energy 193 (2017) 243–262

These days are extracted from the Test Meteorological Year (TMY) file. Once the typical monthly day has been identified, its corresponding energy demand profile (over 24 h) is extracted from !

the overall thermal energy demand profile P I;av g . The procedure is repeated for each month of the season/year. In this way, instead of using the full data set (e.g. 24  365 records for the annual profile or about 24  6  30 records for the seasonal profile of 6 winter months), the first and draft analyses can be based on a smaller and more easy to handle data subset (e.g. 24  12 or 24  6 records for the annual or seasonal profiles, respectively). In short, a statistical approach was developed, in an Excel spreadsheet, which is based on the following rules. The average monthly temperature (T m ) and the average monthly rate of solar radiation ðRm Þ are relatively independent variables. It is almost impossible to find one day that shows exactly the same average temperature and average solar irradiance as the corresponding monthly mean values in the weather data file. Thus, the typical day is assumed as the one that is closest to both parameters, and it is selected from the weather data file by applying the following procedure. Each day of the month (i) in the typical meteorological year (TMY) weather file is characterized by its daily mean Temperature ðT i Þ and daily mean Solar Radiation ðRi Þ.

day i : ðT i ; Ri Þ

ð10Þ

Then DT i and DRi are ordered from the minimum to the maximum. The sum of the serial numbers of the temperature, ðSNDT i Þ, and of the radiation, SNDRi , of the same day is calculated. The typical day, dtyp , is the day that shows the minimum sum:

dtyp ) minðSNDT i þ SNDRi Þ

ð13Þ

If more than one day of the month fits the above criterion (i.e. days that have the same value as the sum of the serial numbers of DT i and DRi ), the typical day is chosen as the day that has the lowest serial number of the temperature ðSNDT i Þ. Thus, in this project, temperature was chosen as the primary variable. An example of this post processing methodology is presented in Table 2 a and b. The January weather data, taken from the Turin (Italy) TMY weather file, was selected, and the typical monthly day for January was defined for Turin according to the above procedure. In other words, the DT i and DRi were calculated and classified, from minimum to maximum, with a given SNDT i and SNDRi respectively. The days of the month ðday i ¼ 1; 2; . . . ; 31Þ were ordered (Table 2b) with their SN DT i and SN DRi . The SNtot was calculated for each day. Day 9 (9th of January), which appears to have the minimum SN tot , is the typical January day in Turin ½dJan;typ ¼ 9 ) minðSN DT i þ SNDRi Þ ¼ 9

5. Example of the application of the methodology to a case study

The temperature and solar radiation variations SN DT i (DT i and

DRi ) from the average monthly values (T m , Rm ) are calculated for each day .

DT i ¼ jT i  T m j½ C

ð11Þ

DRi ¼ jRi  Rm j½Wh=m2 

ð12Þ

A case study was selected in order to apply the methodology, to identify its strengths and weaknesses and to test it. A block of buildings in Turin was selected as the urban district I. The methodology was applied for the entire heating season, and an average overall seasonal heating energy demand profile, with an hourly time resolution, was generated at a block of buildings scale.

Table 2 a. Calculation of temperature and solar radiation variations, b. Selection of the typical monthly day.

254

G. Kazas et al. / Applied Energy 193 (2017) 243–262

Fig. 6. Satellite view of the ‘‘Insula”.

5.2. Step 1- Parametric analysis and database of the energy demand profile formulation

5.1. The block of buildings case study A typical small Italian urban district was selected as the case study. Therefore, in this example, the Insula ðIÞ is represented by a block of buildings that is located at via Barrili 5, Turin, Italy. This is a rectangular block of buildings and it is made up of twelve buildings:

I ¼ fB1 ; B2 ; B3 ; B4 ; . . . ; B12 g

ð14Þ

All the buildings are multi-family residential buildings, and most of them are connected to the Turin district heating network. The buildings have similar construction and architectural features, while there are some differences in their height and in the number of floors. They were all built according to the same Italian regulations. The shape, configuration and the other thermal parameters of the buildings in the Insula are quite similar. It was therefore possible to adopt just one building as a Reference Building ðRB;1 Þ. For the sake of brevity, and to simplify the example as much as possible, just one reference building was assumed to be representative of the entire set of buildings in the Insula. Another reason for this simplified choice was linked to the fact that the measured data (which were necessary for the model calibration, as will be explained in Section 5.2.1) were only available for one building of the Insula. The reference building ðRB;1 Þ is a real building in the block. It was simulated in detail by means of a comprehensive white box approach, where each room was an individual thermal zone. The reference building, which can be defined as an Example Reference Building [55], was constructed with a technology that is typical of Turin. It is connected to the Turin district heating network through a heat exchanger, which has a nominal capacity of 150 kW. A satellite view of the block of buildings and the geometry of the Reference Building are shown in Figs. 6 and 7, respectively. The main geometrical parameters of the building are summarised in Table 3.

5.2.1. Reference building modelling and calibration The white-box model of the Reference Building RB;1 was developed in EnergyPlus v.7.0.0 [61], using OpenStudio Plug-in [62] for the geometrical design. The geometrical characteristics are based on the energy performance certificate and on-site inspections of the building. The typical floor consists of two flats, and each flat has five rooms. The walls are made of brick, with no thermal insulation. The hydronic heating system of the building consists of vertical columns and of cast-iron radiator units. The occupants’ density, lighting and electrical equipment values of the ‘‘base case scenario” are 0.04 people/m2, 3.88 W/m2 and 5.38 W/m2, respectively. The air infiltration rate was set to 0.3 ACH [63]. This building and the corresponding energy model were selected as prototypes for the following reasons:
A detailed and reliable set of sub-hourly metered heating load profiles were available for this building (from the district heating supplier), covering an entire heating season,
An initial version of the building model had already been developed by Monetti et al. [63], accordingly to Italian and international regulations and standards within a study about the effect of thermostatic radiator valves on district heating loads [64]. A calibration approach was developed to evaluate the reliability of the Reference Building model [65]. The Reference Building model was simulated in EnergyPlus v.7.0.0 [62], using the weather data file for the 2011–2012 heating season. The seasonal heating profile was calibrated with real metered data collected during the same period of time (these data were only available for one building located in the Insula. The building is highlighted in

G. Kazas et al. / Applied Energy 193 (2017) 243–262

255

Fig. 7. OpenStudio Plug-in version of the reference building model.

Table 3 Geometrical data of the reference building. Construction year

Storeys

Total gross area (m2)

Number of flats

1933

5

1000

12

This amount of heating energy was added to the morning peak of each day, and the building model was calibrated with the metered data.

Fig. 6. The simulated profile had an hourly time step, while the metered data from the building substation of the district heating network had a time step of between 4–6 min (sometimes more than 10 min, and some of the data were missing). In order to compare measured data with simulated data, they both need to have the same time step. The time steps of the measured data were modified in order to obtain an hourly time step. This was achieved by calculating the average hourly value of the measured data. A comparison of the simulated and the measured demand profiles was carried out with the same time step. During the first assessment of the results, a difference was observed between the morning peak of the simulated and measured profiles. The morning peak is the heating load that is provided to the building during the first period of operation in the morning (5:00–7:00am). The building model, the schedules and the real energy system configuration were analysed and assessed, and the reason for this incongruence was found to be an underestimation of the simulated morning peak, in comparison to the measured data peak. This was due to the water content in the heating loop. The real system needs more heating energy in the morning to heat the water contained in the circulation system. The calculation of the thermal energy needed to re-heat the water contained in the heating system was computed by means of the following heat capacity equation

Q ¼ mc  cp  DT½kW h

temperature of each month (Tfinal = avgTOct, avgTNov, ..., avgTApr) in the calibration. These average temperatures were estimated from metered data from 2011 to 2012.

ð15Þ

where
mc is the water content, which was defined as 5 Litres per kW of installed power capacity according to regional regulations. This content was multiplied by the maximum heating capacity of the season in order to calculate the water content of the system.
cp is the specific heat of the water and is 4.186 kJ/kg K.
DT is the temperature difference between the temperature of the water when the heating system is off (always 20 °C) and the temperature that the water is reaching each morning. This temperature varies from day to day and from month to month as a function of the outdoor climate (climatic regulation). This temperature was assumed to be the average inlet water

5.2.2. Parametric analysis of the reference building model A parametric analysis was carried out according to the developed methodology. The parametric analysis in the Reference Building model ðRB;1 Þ included various scenarios, and resulted in a significant number of heating energy demand profiles with hourly time steps. These heating demand profiles, normalized by the Reference Building model conditioned floor area ðARB;1 ;cond: Þ, were used to create a database of the normalized heating energy demand profiles that cover various internal load and other thermal parameter scenarios. Profiles for the rest of the buildings in the block may be extracted from this database through a random selection in order to generate all the average heating demand profiles for the entire block of buildings ðIÞ. The random selection of the profiles from the database was chosen in order to generalize this methodology, which may be used for any block of newly constructed or already existing urban district. The various scenarios that were developed on the basis of the methodology presented in 4.2.2 were introduced as inputs into the EnergyPlus parametric tool [61] (Parametric:SetValueForRun and Parametric:RunControl). The parametric analysis scenarios of the Reference Building model were simulated in EnergyPlus [61]. Since the considered building is located in Turin (Italy), it was simulated with the Turin IWEC TMY weather file. The variables that were used for the development of the scenarios, and which were investigated through the parametric analysis, were the relative position of the building ðSs;k Þ, the internal loads, such as the occupant’s density, lighting and electrical equipment loads ðSl;m Þ, and the infiltration rates ðSv ;g Þ. Four shading scenarios were considered, and the various relative positions of the buildings in the block of buildings I were covered. The orientation of the Reference Building was chosen as a base case scenario. The base case scenario ðSs;1 Þ, was 0°, and the main façade was North-East orientated. The other orientation scenarios were created by adding 90° to the base case scenario (Table 4). Three infiltration rate scenarios and a parametric analysis range of ±20% were considered, in comparison to the base case infiltration rate ðSv ;1 ¼ 0:3ACHÞ. Three main internal load scenarios and

256

G. Kazas et al. / Applied Energy 193 (2017) 243–262

Table 4 Parametric analysis scenarios. Scenarios Shading (°) Infiltration rate (ACH) ðSl;m Þ Occupants’ density (people/m2) Lighting loads (W/m2) Electrical equip. (W/m2)

Ss;1 : 0 (NE) Sv ;1 : 0.28 Light use 0.03 2.92 4.035

Ss;2 : 90 (SE) Sv ;3 : 0.35 Typical use 0.04 3.88 5.38

a distribution of between 25% and +25% were analysed, in comparison to the base case scenario ðSl;1 : occupants density : W W 0:04 people ; lighting : 3:88 m2 and electrical equipment : 5:38 m2 Þ. m2 Flats that were considered empty were also tested for approximately 25% of the building area. The internal load scenarios Sl;m that were influenced directly by the occupants’ behaviour were simulated using a percentage distribution of the floor area. Twenty-five internal load scenarios, Sl;m , were simulated for each Ss;k and Sv ;g This was done due to the diversity of the occupants’ behaviour and of the energy performance at the two different scales, that is, the difference between the various buildings at the block of building scale and between the floors and the apartments at the building scale. A total of 300 scenarios were developed in the parametric analysis, 75 scenarios for each Ss;k orientation. EnergyPlus v.7.0.0 [61] was used for the simulation of the parametric analysis. The parametric analysis led to a total of 300 seasonal heating rate profiles, with an hourly time step, for the Reference Building. These heating energy demand profiles were normalized according to the conditioned floor area of the Reference Building model. The heating energy demand profiles were expressed in kW/m2. This normalization of the heating rate profiles was carried out in order to cover the variety areas and heights of the buildings in the block. The normalized heating energy demand profiles were used to construct the D database. A district scale heating demand profiles generation is carried out in the next section using this database.

5.3. Step 2 – Aggregation method and generation of the average heating demand profile of a block of buildings The Methodology that was developed in Section 4.3 was applied, and an average heating energy demand profile was generated for a block of buildings (see Table 5). The database of the normalized heating energy demand profiles, which is equal to the D1 subdatabase (n = 1 reference building), consisted of 300 representative normalized heating energy demand profiles of the block of buildings. Table 5 Storeys in the buildings and conditioned floor areas of the block of buildings. Ss;k

Bx

Storey

ABx ;cond ðm2 Þ

Ss;1

B1 B2 B3 B4 B5 B6 B7 B8 B9 B1 0 B1 1 B1 2

6 5 5 6 8 3 6 5 6 7 10 and 9 2

676.20 728.00 661.50 1600.20 2324.00 579.60 848.40 558.25 809.34 1053.50 1962.93 307.16

Ss;2 Ss;3

Ss;4

Ss;3 : 180 (SW) Sv ;3 : 0.42 Intensive use 0.05 4.85 6.725

Ss;4 :270(NW) Empty 0.00 0.00 0.00

8 9 for k ¼ 1;1 6 g 6 3;1 6 m 6 25;C 1;Ss;1 : fp! ðR1 ;S1;g;m Þ g > > > > > > > > > > > < for k ¼ 2;1 6 g 6 3;1 6 m 6 25;C 1;Ss;2 : fp! ðR1 ;S2;g;m Þ g > = D ¼ D1 : ;a ¼ 4 > for k ¼ 3;1 6 g 6 3;1 6 m 6 25;C 1;Ss;3 : fp! ðR1 ;S3;g;m Þ g > > > > > > > > > > > : ; for k ¼ 4;1 6 g 6 3;1 6 m 6 25;C 1;Ss;4 : fp! ðR1 ;S4;g;m Þ g ð16Þ Four clusters were therefore created in the database, according to the four orientation scenarios ða ¼ 4Þ that were identified for the block of buildings. The number of storeys and their relative positions were identified through on-site inspections. A shading scenario (cluster) was matched to each building. A normalized heating profile was then extracted from the cluster. The selection was carried out in a random way. This random approach was adopted to cover the uncertainty connected to the behaviour of the various occupants, and to create a general approach that could be applied not only to this case study, but also to any block of buildings. Each normalized heating demand profile was multiplied by the conditioned floor area of each building, ABx ;cond . Each building in the block was matched to a heating demand profile for the first thermal energy demand profile, and then all the profiles were summed. The sum was equal to the heating energy demand profile of the total block of buildings that had been generated. The same process was repeated 100 times. Thus, 100 heating energy demand profiles, with an hourly time resolution, were generated for the current block of buildings (12 buildings).An average heating profile was calculated from these profiles in order to be representative of the block. The choice of j = 100 times was made to investigate the effect of enlarging the profile sample. 5.4. Analysis of the results DiDeProM was applied to a small urban district, but with some assumptions, in order to generate one overall average heating demand profile at a block of buildings scale in Turin, Italy. 5.4.1. Calibration After the modelling of the reference building, RB;1 , a calibration was carried out. Figs. 8 and 9 indicate the results of the calibration for three days (11th, 12th and 13th) of November and of December. The following profiles are presented in the Figs. 8 and 9:
A ‘‘real heating rate profile”, which is the metered heating profile for 2012 for the reference building
A simulated profile (before calibration), which is the profile simulated in EnergyPlus according to Step 1 in DiDeProM
A calibrated DiDeProM profile, which is the calibrated simulated profile after the modification of the morning peak, as described in 5.2.1. In these examples, the differences in the morning energy peak between the metered profiles and the uncalibrated simulated

G. Kazas et al. / Applied Energy 193 (2017) 243–262

257

Fig. 8. Calibration results for 11th, 12th and 13th of November.

Fig. 9. Calibration results for 11th, 12th and 13th of December.

profiles underline some differences in behaviour between the real building and the simulated model. The differences pertain to the different operations of the district heating system. The modelled heating system does not take into account the heating network and in particular the district heating network configuration. Thus, in reality, since the system is turned off during the night time, when the system starts at 5 o’clock in the morning it needs a huge amount of energy to heat all the water contained in the pipe and to reach the set supply water temperature of the system. This amount of energy was calculated and added to the morning peak. The calibrated profiles fitted well with the metered profiles. The Root Mean Square Error (RMSE) was chosen to evaluate the calibration results. The result of this criterion indicated that the calibrated simulated heating profiles were close to the real metered profiles. The RMSErrors calculated for the uncalibrated simulated heating profiles and the calibrated heating profiles are compared in Table 6. The RMSErrors for November, December and March (representative months of the heating season for autumn, winter and spring) are presented in Table 6. The results

Table 6 RMSE evaluation of the calibration.

November December March

Uncalibrated simulation RMSE (kW)

Calibrated simulation RMSE (kW)

4.03 4.02 4.59

1.97 2.86 3.39

show that the RMSErrors of the calibrated simulated heating profiles were lower than the RMSErrors of the uncalibrated simulated heating profiles and that the heating rate in kW was very low.

RMSEcalibratedsim < RMSEuncalinratedsim

ð17Þ

5.4.2. Parametric analysis After the modelling and calibration of the reference building, RB;1 , 300 scenarios were developed and simulated in a parametric analysis. This parametric analysis led to the same number of seasonal heating demand profiles, with an hourly time step normalized by the RB;1 conditioned floor area. These normalized profiles, which are the outcomes of the parametric analysis, were summed in order to calculate the total heating energy demand of the heating season (Fig. 10). The total seasonal heating demand is presented in Fig. 10 where it is classified into Shading Scenario clusters to form the normalized heating demand profile database. The parametric analysis pattern of the results is also clear from the representation in Fig. 10. The percentage difference in the total seasonal heating demand between the two extreme scenarios of the parametric analysis is approximately 18%. This difference can be considered small, compared to the variation in the parametric analysis variables that were taken into account. This difference is due to the size of the building, which is a 5-storey building (about 706 m2 conditioned floor area), and to the very detailed physical model. In order to assess the impact of the parametric analysis on the daily heating profiles, the typical monthly day heating profiles

258

G. Kazas et al. / Applied Energy 193 (2017) 243–262

Fig. 10. Normalized total seasonal heating demand database.

were calculated. The average, minimum and maximum profiles of all the parametric analysis scenarios were calculated for each of the typical day profiles. A typical December day and a typical March day profile are presented in Figs. 11 and 12, respectively. These profiles represent a characteristic winter and a characteristic spring profile. The difference in the amount of energy and the different morning peaks for these two completely difference seasons were to be expected, considering the regulations, schedule and performance of the system.

Fig. 11. Min, avg and max normalized heating demand profiles for a typical day in December.

Fig. 12. Min, avg and max normalized heating demand profiles for a typical day in March.

5.4.3. Heating demand profile generation for the block of buildings Twelve normalized seasonal heating demand profiles were generated for the given block of buildings using the normalized profiles database and step 2 in DiDeProM. These normalized heating demand profiles were multiplied by the conditioned floor area of each building. The seasonal heating demand profiles of the 12 buildings were summed, and the result was an overall seasonal heating demand profile for the block of buildings (Fig. 13). This seasonal heating demand profile has an hourly time step. As previously mentioned, in order to make the methodology more general, the heating demand profile generation for the block of buildings was repeated times. In the current paper, j was chosen equal to 100 times. Initially, 20 seasonal heating demand profiles were generated from the database for the block of buildings. This sample was enlarged by 20 profiles each time, until the sample consisted of 100 heating demand profiles for the block of buildings. The total sample of the block of buildings is indicated in Fig. 14. This figure includes 100 points that represent the overall seasonal heating demand for the 100 heating demand profiles selected from the database.

Fig. 13. Overall heating demand profile for the block of buildings.

259

G. Kazas et al. / Applied Energy 193 (2017) 243–262

Seasonal Heang Demand (MWh)

1120 1110 1100 1090 1080 1070 1060 1050 1040 1030 0

20

40

60

80

100

Number of profile generaon Fig. 14. Sample of the heating demand profiles for 100 blocks of buildings Seasonal heating demand.

Heang energy (MWh)

After the application of the current methodology, the dispersion between the maximum and minimum seasonal heating demand values was far less than the same dispersion at the building scale. The seasonal heating demand of the block of buildings presents a percentage dispersion of approximately 7%, in comparison to the

1100

0,40%

1090

0,35% 0,30%

1080

0,25% 1070 0,20% 1060 0,15% 1050

0,10%

1040 1030

0,05%

Dispersion of total seasoal energy demand Absolute percentage variation of the samples average 20

40

60

Absolute percentage variaon

18% of the building scale. In order to underline the small dispersion, and thus the reliability of the average heating demand profile, a statistical interpretation of the various profile samples was carried out. Fig. 15 indicates the dispersion of the 5 samples of the heating profiles of the block of buildings, starting from the sample of 20 profiles and enlarging it to 40, 60, 80 and 100 profiles. As can be seen, the median is almost stable and 50% of the values of each sample are in almost the same heating rate range. The absolute percentage variation of the average of the samples is also presented in the same figure. This is the absolute variation of the average of one sample compared to the average of the previous smaller sample (Absolute percentage variation 40–20, 60– 40, 80–60 and 100–80). This indicator declined as the sample was enlarged, and almost reached a zero percentage variation when the sample of the average of 100 heating demand profiles was compared with the sample of the average of 80 heating demand profiles. Therefore, the generation of the heating demand profiles for the block of 100 buildings can be considered a good sample to calculate the average seasonal heating demand profile for storage technologies and demand side management investigations. The last outcome of the current methodology is the cumulative heating load curve. This cumulative curve indicates the heating

1130

0,00% 80

100

Profile samples Fig. 15. Dispersion of the total seasonal heating demand for j = 20, 40, 60 80 and 100 and absolute percentage variation of the samples average.

Fig. 16. Cumulative heating load curves at a block of buildings scale.

260

G. Kazas et al. / Applied Energy 193 (2017) 243–262

Table 7 Typical days for the different months of the heating season for Turin, Italy.

Typical day Dry outdoor temperature [°C] Solar radiation [Wh/m2]

October

November

December

January

February

March

April

24 Oct 11.24 69.96

09 Nov 5.91 78.29

08 Dec 3.08 40.83

09 Jan 1.33 53.04

09 Feb 5.91 78.29

25 Mar 9.10 118.58

04 Apr 13.26 79.58

load pattern for the block of buildings during a heating season. The average heating rate of the heating demand profiles for the block of 100 buildings is presented in Fig. 16. The behaviour of the system was as expected. It is clear that if the cumulative curve and the average heating demand profile are used, a strategy can be developed for thermal energy storage and renewable energy technology integration at a district scale. 5.4.4. Selection of the typical days The final step of the current methodology is the postprocessing. This activity includes the selection of typical days from Turin’s TMY weather file. The typical days for Turin (Italy) are presented in Table 7 according to the methodology that has been developed in the current paper. Profiles of the district for the typical days are useful because they allow to study the dynamic energy performance of the district by means of simple daily load profiles. The final and practical aim of the DiDeProM model is in fact to provide reliable and hourly thermal load profiles of aggregation of buildings in order to study demand side management strategies, integration of thermal energy storages and renewable energy technologies at a district scale. A promising application of the model will also to provide detailed hourly data thermal load profiles to district heating companies for refining the design of new pipelines. 6. Conclusions The development of strategies that can be used for an integrated demand side management using thermal energy storage technologies at an urban district scale has here been studied as part of the European CI-NERGY project. The urban district scale can vary, and its choice depends on the interest of the engineers and urban planners. In order to develop demand side managements strategies, and to apply and integrate active thermal energy storage technologies, it is necessary to generate thermal energy demand profiles at an urban district scale. These profiles must have a very detailed time resolution. The thermal energy demand profiles of an urban district, with hourly time resolution, could be used to develop a demand side management strategy integrated with thermal energy storage technologies. In the current paper, a bottom-up engineering model has been developed for a thermal energy demand profile, with an hourly time step generation, at an urban district scale. This model, which is called DiDeProM, consists of two main steps: (a) the development of a database that consists of normalized thermal energy demand profiles at a building scale and (b) the generation of an overall thermal energy demand profile at an urban district scale. The model is able to satisfy two important research issues that were identified in the state of art review. The development of an aggregation method, which could cover the uncertainties of a district (parametric analysis), and would be sufficiently precise and general to be suitable for already existing or newly constructed urban districts. The second and most important issue was to answer the question on how district demand profiles, with a very detailed time resolution, could be generated for use in energy systems and storage technology applications. The result of the current work is an average urban district thermal energy demand profile

with an hourly time step. In case of districts less homogeneous than the one presented as a case study, the user of DiDeProm should perform the parametric analysis on a larger number of reference buildings that will be selected from the district based on a classification and clustering of the actual buildings of the district. The limitation of this modelling approach can be found in case no clusters can be identified to classify the buildings of a district. In this case, this method can be classified as a ‘‘brute force” (see paragraph 2.2.1) since each building has to be modelled in detail. The next step, in order to generalize DiDeProM, will be to assess and include input on dynamic occupancy behaviour. The behaviour of occupants is the most important variable in the current study. Stochastic models for the prediction of occupants’ behaviour are currently under development [66]. The integration of these models with DiDeProM could help to generalize the model and make it more precise. As previously mentioned, detailed district thermal energy demand profiles could be used to carry out an optimization of the energy demand at an urban district scale. This optimization could be carried out by applying demand side management strategies [67], using thermal energy storage technologies [68]. Detailed thermal energy demand profiles could therefore be used to create a model and a strategy for the integration of thermal energy storage technologies as well as for the development of a decentralized thermal energy storage and renewable energy technology network at an urban district scale. The research consortium of the CI-NERGY project is working to extract building data that are necessary to run the DiDeProM tool from a detailed georeferenced information database at the district level. New case studies will therefore be available through the query of the database. In particular, an application to the Meyrin district in the Canton of Geneva is going to be developed. In this case, actual measurements form the district heating substations of the various buildings of the district are available for various heating seasons and may be used to further improve the reliability of the DiDeProm model. Acknowledgements The authors gratefully acknowledge the European Commission for having provided the financial support for the present research as part of the FP7-PEOPLE-2013 Marie Curie ‘‘CI-NERGY” Initial Training Network project with Grant Agreement Number 606851. The authors would also like to thank Valentina Monetti for sharing the first version of the reference building model. References [1] Kikegawa Y, Genchi Y, Kondo H, Hanaki K. Impacts of city-block-scale countermeasures against urban heat-island phenomena upon a building’s energy-consumption for air-conditioning. Appl Energy 2006;83:649–68. http://dx.doi.org/10.1016/j.apenergy.2005.06.001. [2] De Rosa M, Bianco V, Scarpa F, Tagliafico La. Heating and cooling building energy demand evaluation; a simplified model and a modified degree days approach. Appl Energy 2014;128:217–29. http://dx.doi.org/10.1016/j. apenergy.2014.04.06. [3] European Parliament, Council E. Directive 2010/31/EU of the european parliament and of the council of 19 May 2010 on the energy performance of buildings (recast); 2010. [4] Jain RK, Smith KM, Culligan PJ, Taylor JE. Forecasting energy consumption of multi-family residential buildings using support vector regression: Investigating the impact of temporal and spatial monitoring granularity on

G. Kazas et al. / Applied Energy 193 (2017) 243–262

[5]

[6]

[7]

[8]

[9] [10]

[11]

[12]

[13]

[14] [15]

[16]

[17]

[18]

[19]

[20] [21]

[22] [23]

[24] [25]

[26]

[27]

[28]

[29]

[30]

[31]

performance accuracy. Appl Energy 2014;123:168–78. http://dx.doi.org/ 10.1016/j.apenergy.2014.02.057. Foucquier A, Robert S, Suard F, Stéphan L, Jay A. State of the art in building modelling and energy performances prediction: a review. Renew Sustain Energy Rev 2013;23:272–88. http://dx.doi.org/10.1016/j.rser.2013.03.004. Fonseca JA, Schlueter A. Integrated model for characterization of spatiotemporal building energy consumption patterns in neighborhoods and city districts. Appl Energy 2015;142:247–65. http://dx.doi.org/10.1016/j. apenergy.2014.12.068. Reinhart CF, Davila CC. Urban building energy modeling e A review of a nascent fi eld. Build Environ 2016;97:196–202. http://dx.doi.org/10.1016/j. buildenv.2015.12.001. Mckenna R, Merkel E, Fichtner W. Energy autonomy in residential buildings: a techno-economic model-based analysis of the scale effects. Appl Energy 2017;189:800–15. http://dx.doi.org/10.1016/j.apenergy.2016.03.062. Dilay Kesten Erhart. CINERGY Project. Marie Curie Cinergy Proj; 2015. [accessed April 15, 2016]. Bertrand A, Mastrucci A, Schüler N, Aggoune R, Maréchal F. Characterisation of domestic hot water end-uses for integrated urban thermal energy assessment and optimisation. Appl Energy 2017;186:152–66. http://dx.doi.org/10.1016/j. apenergy.2016.02.107. Hargreaves A, Cheng V, Deshmukh S, Leach M, Steemers K. Forecasting how residential urban form affects the regional carbon savings and costs of retrofitting and decentralized energy supply. Appl Energy 2017;186:549–61. http://dx.doi.org/10.1016/j.apenergy.2016.02.095. Best RE, Flager F, Lepech MD. Modeling and optimization of building mix and energy supply technology for urban districts. Appl Energy 2015;159:161–77. http://dx.doi.org/10.1016/j.apenergy.2015.08.076. Fracastoro GV, Serraino M. A methodology for assessing the energy performance of large scale building stocks and possible applications. Energy Build 2011;43:844–52. http://dx.doi.org/10.1016/j.enbuild.2010.12.004. Choudhary R. A probalistic model for assessing energy consumption of the non-domestic building stock. Build. Simul. 2011:14–6. Cheng V, Steemers K. Modelling domestic energy consumption at district scale: a tool to support national and local energy policies. Environ Model Softw 2011;26:1186–98. http://dx.doi.org/10.1016/j.envsoft.2011.04.005. Swan LG, Ugursal VI. Modeling of end-use energy consumption in the residential sector: a review of modeling techniques. Renew Sustain Energy Rev 2009;13:1819–35. http://dx.doi.org/10.1016/j.rser.2008.09.033. Kavgic M, Mavrogianni A, Mumovic D, Summerfield A, Stevanovic Z, DjurovicPetrovic M. A review of bottom-up building stock models for energy consumption in the residential sector. Build Environ 2010;45:1683–97. http://dx.doi.org/10.1016/j.buildenv.2010.01.021. Hong S-M, Paterson G, Burman E, Steadman P, Mumovic D. A comparative study of benchmarking approaches for non-domestic buildings: Part 1 – Topdown approach. Int J Sustain Built Environ 2013;2:119–30. http://dx.doi.org/ 10.1016/j.ijsbe.2014.04.001. Ren Z, Paevere P, McNamara C. A local-community-level, physically-based model of end-use energy consumption by Australian housing stock. Energy Policy 2012;49:586–96. http://dx.doi.org/10.1016/j.enpol.2012.06.065. Runkler TA. Data Analytic. Springer Vieweg; 2012. Choudhary R. Energy analysis of the non-domestic building stock of Greater London. Build Environ 2012;51:243–54. http://dx.doi.org/10.1016/j. buildenv.2011.10.006. Tian W, Choudhary R. Energy use of buildings at urban scale: a case study of London school buildings. Build Simul 2011:14–6. Tian W, Choudhary R. A probabilistic energy model for non-domestic building sectors applied to analysis of school buildings in greater London. Energy Build 2012;54:1–11. http://dx.doi.org/10.1016/j.enbuild.2012.06.031. Yamaguchi Y, Suzuki Y, Choudhary R, Booth A, Shimoda Y. Urban-scale energy modelling of food supermarket considering uncertainty. IBPSA; 2013. Mastrucci A, Baume O, Stazi F, Leopold U. Estimating energy savings for the residential building stock of an entire city: a GIS-based statistical downscaling approach applied to Rotterdam. Energy Build 2014;75:358–67. http://dx.doi. org/10.1016/j.enbuild.2014.02.032. Julia S, Davila C, Christoph FR. Validation of a Bayesian-based method for defining residential archetypes in urban building energy models. Energy Build 2017;134:11–24. http://dx.doi.org/10.1016/j.enbuild.2016.10.050. Kohler M, Blond N, Clappier A. A city scale degree-day method to assess building space heating energy demands in Strasbourg Eurometropolis (France). Appl Energy 2016;184:40–54. http://dx.doi.org/10.1016/j. apenergy.2016.09.075. Ihm P, Krarti M. Design optimization of energy efficient residential buildings in Tunisia. Build Environ 2012;58:81–90. http://dx.doi.org/10.1016/j. buildenv.2012.06.012. Hamdy M, Hasan A, Siren K. Applying a multi-objective optimization approach for Design of low-emission cost-effective dwellings. Build Environ 2011;46:109–23. http://dx.doi.org/10.1016/j.buildenv.2010.07.006. Ekpenyong UE, Zhang J, Xia X. Mathematical modelling for the social impact to energy efficiency savings. Energy Build 2014;84:344–51. http://dx.doi.org/ 10.1016/j.enbuild.2014.08.019. Pisello AL, Taylor JE, Xu X, Cotana F. Inter-building effect: Simulating the impact of a network of buildings on the accuracy of building energy performance predictions. Build Environ 2012;58:37–45. http://dx.doi.org/ 10.1016/j.buildenv.2012.06.017.

261

[32] Hachem C, Athienitis A, Fazio P. Evaluation of energy supply and demand in solar neighborhood. Energy Build 2012;49:335–47. http://dx.doi.org/10.1016/ j.enbuild.2012.02.021. [33] Ward RM, Choudhary R. A bottom-up energy analysis across a diverse urban building portfolio: retrofits for the buildings at the Royal Botanic Gardens, Kew, UK. Build Environ 2014;74:132–48. http://dx.doi.org/10.1016/j. buildenv.2013.12.018. [34] Pittam J, O’Sullivan PD, O’Sullivan G. Stock aggregation model and virtual archetype for large scale retro-fit modelling of local authority housing in Ireland. Energy Proc 2014;62:704–13. http://dx.doi.org/10.1016/ j.egypro.2014.12.434. [35] Jones PJ, Lannon S, Williams J. Modelling building energy use at urban scale. Brazil: Seventh Int. IBPSA Conf. Rio Janeiro; 2001. p. 175–80. [36] Clarke J, Johnstone C, Kondratenko I, Lever M, McElroy L, Prazeres L, et al. Using simulation to formulate domestic sector upgrading strategies for Scotland. Energy Build 2004;36:759–70. http://dx.doi.org/10.1016/j. enbuild.2004.01.034. [37] Mata É, Sasic Kalagasidis A, Johnsson F. Building-stock aggregation through archetype buildings: France, Germany, Spain and the UK. Build Environ 2014;81:270–82. http://dx.doi.org/10.1016/j.buildenv.2014.06.013. [38] Mata É, Kalagasidis AS, Johnsson F. A modelling strategy for energy, carbon, and cost assessments of building stocks. Energy Build 2013;56:100–8. http:// dx.doi.org/10.1016/j.enbuild.2012.09.037. [39] Mata É, Sasic Kalagasidis A, Johnsson F. Energy usage and technical potential for energy saving measures in the Swedish residential building stock. Energy Policy 2013;55:404–14. http://dx.doi.org/10.1016/j.enpol.2012.12.023. [40] Tuominen P, Holopainen R, Eskola L, Jokisalo J, Airaksinen M. Calculation method and tool for assessing energy consumption in the building stock. Build Environ 2014;75:153–60. http://dx.doi.org/10.1016/j.buildenv.2014.02.001. [41] Wang Q, Holmberg S. A methodology to assess energy-demand savings and cost effectiveness of retrofitting in existing Swedish residential buildings. Sustain Cities Soc 2015;14:254–66. http://dx.doi.org/10.1016/j. scs.2014.10.002. [42] Shorrock LD, Dunster JE. The physically-based model B R E H O M E S and its use in deriving scenarios for the energy use and carbon dioxide emissions of the UK housing stock 1997;25:1027–37. [43] Huang YJ, Brodrick J. A Bottom-Up Engineering Estimate of the Aggregate Heating and Cooling Loads of the Entire U. S. Building Stock; 2000. [44] Hens H, Verbeeck G, Verdonck B. Impact of energy efficiency measures on the CO 2 emissions in the residential sector, a large scale analysis. Energy Build 2001;33:275–81. [45] Shimoda Y, Fujii T, Morikawa T, Mizuno M. Evaluation on Residential Energy Efficiency Programs Using the City-Scale End-Use Simulation Model 2004:249–59. [46] Shimoda Y, Fujii T, Morikawa T, Mizuno M. Residential end-use energy simulation at city scale. Build Environ 2004;39:959–67. http://dx.doi.org/ 10.1016/j.buildenv.2004.01.020. [47] Shimoda Y, Asahi T, Taniguchi A, Mizuno M. Evaluation of city-scale impact of residential energy conservation measures using the detailed end-use simulation model. Energy 2007;32:1617–33. http://dx.doi.org/10.1016/j. energy.2007.01.007. [48] Yamaguchi Y, Shimoda Y, Mizuno M. Proposal of a modeling approach considering urban form for evaluation of city level energy management. Energy Build 2007;39:580–92. http://dx.doi.org/10.1016/j. enbuild.2006.09.011. [49] Farahbakhsh H, Ugursal VI, Fung AS. A residential end-use energy consumption model for Canada. Int J Energy Res 1998;22:1133–43. [50] Chen S, Yang W, Yoshino H, Levine MD, Newhouse K, Hinge A. Definition of occupant behavior in residential buildings and its application to behavior analysis in case studies. Energy Build 2015. http://dx.doi.org/10.1016/j. enbuild.2015.06.075. [51] Wei S, Jones R, de Wilde P. Driving factors for occupant-controlled space heating in residential buildings. Energy Build 2014;70:36–44. http://dx.doi. org/10.1016/j.enbuild.2013.11.001. [52] Langevin J, Wen J, Gurian PL. Simulating the human-building interaction: development and validation of an agent-based model of office occupant behaviors. Build Environ 2015;88:27–45. http://dx.doi.org/10.1016/j. buildenv.2014.11.037. [53] Virote J, Neves-Silva R. Stochastic models for building energy prediction based on occupant behavior assessment. Energy Build 2012;53:183–93. http://dx. doi.org/10.1016/j.enbuild.2012.06.001. [54] Hsu D. Comparison of integrated clustering methods for accurate and stable prediction of building energy consumption data. Appl Energy 2015;160:153–63. http://dx.doi.org/10.1016/j.apenergy.2015.08.126. [55] Corgnati SP, Fabrizio E, Filippi M, Monetti V. Reference buildings for cost optimal analysis: method of definition and application. Appl Energy 2013;102:983–93. http://dx.doi.org/10.1016/j.apenergy.2012.06.001. [56] Ballarini I, Corgnati SP, Corrado V. Use of reference buildings to assess the energy saving potentials of the residential building stock: the experience of TABULA project. Energy Policy 2014;68:273–84. http://dx.doi.org/10.1016/j. enpol.2014.01.027. [57] Davila CC, Reinhart CF, Bemis JL. Modeling Boston : a work fl ow for the efficient generation and maintenance of urban building energy models from existing geospatial datasets. Energy 2016;117:237–50. http://dx.doi.org/ 10.1016/j.energy.2016.10.057.

262

G. Kazas et al. / Applied Energy 193 (2017) 243–262

[58] Pisello AL, Asdrubali F. Human-based energy retrofits in residential buildings: a cost-effective alternative to traditional physical strategies. Appl Energy 2014;133:224–35. http://dx.doi.org/10.1016/j.apenergy.2014.07.049. [59] Saelens D, Parys W, Baetens R. Energy and comfort performance of thermally activated building systems including occupant behavior. Build Environ 2011;46:835–48. http://dx.doi.org/10.1016/j.buildenv.2010.10.012. [60] European Commission. Population and housing census. Eurostat; 2016. [accessed June 23, 2016]. [61] USD of E. EnergyPlus Energy Simulation Software; 2015. [accessed June 23, 2016]. [62] Alliance for Sustainable Energy. OpenStudio n.d. [accessed June 23, 2016]. [63] Monetti V, Fabrizio E, Filippi M. Influence of different temperature control patterns through TRV on district heating loads. In: Li A, Zhu Y, Li Y, editors. Proc 8th Int Symp Heating, Vent Air Cond, vol. 262, Berlin, Heidelberg: Springer, Berlin Heidelberg; 2014, p. 251–8. http://dx.doi.org/10.1007/978-3642-39581-9. [64] Monetti V, Fabrizio E, Filippi M. Impact of low investment strategies for space heating control: application of thermostatic radiators valves to an old residential building. Energy Build 2015;95:202–10. http://dx.doi.org/ 10.1016/j.enbuild.2015.01.001. [65] Fabrizio E, Monetti V. Methodologies and advancements in the calibration of building energy models. Energies 2015;8:2548–74. http://dx.doi.org/10.3390/ en8042548.

[66] Wilke U, Haldi F, Scartezzini J-L, Robinson D. A bottom-up stochastic model to predict building occupants’ time-dependent activities. Build Environ 2013;60:254–64. http://dx.doi.org/10.1016/j.buildenv.2012.10.021. [67] Gellings CW, Smith WM. Integrating demand-side management into utility planning. Proc IEEE 1989;77:908–18. [68] Arteconi A, Hewitt NJ, Polonara F. State of the art of thermal storage for demand-side management. Appl Energy 2012;93:371–89. http://dx.doi.org/ 10.1016/j.apenergy.2011.12.045.

Georgios Kazas was born in Tessaloniki on 30 July, 1985. He graduated in Civil Engineering at the Polytechnic Faculty of the Democritus University of Thrace in Greece, and then did an MSc in Sustainable Energy, Technology and Management at Brunel University. In 2014, he started at the Politecnico di Torino as an Early Stage Researcher, as part of the Marie Curie CI-NERGY Project, and as a PhD student in Energetics in the field of ‘‘Energy supply and demand management through energy storage and demand side management”. He passed away on July 16, 2016 while going to his office. The co-authors would like to dedicate this paper to his memory, and together with all the people of the research consortium would like to express their sorrow for his premature departure.