Improvements of the operational rating system for existing residential buildings

Applied Energy 193 (2017) 112–124

Contents lists available at ScienceDirect

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

Improvements of the operational rating system for existing residential buildings Jaewook Jeong a, Taehoon Hong a,⇑, Changyoon Ji a, Jimin Kim a, Minhyun Lee a, Kwangbok Jeong a, Choongwan Koo b a b

Department of Architectural Engineering, Yonsei University, 50 Yonsei-ro, Seodaemun-gu, Seoul 03722, Republic of Korea Department of Building Services Engineering, Hong Kong Polytechnic University, Hung Hom, Kowloon, Hong Kong

h i g h l i g h t s
Building Energy Consumption Certification (BECC) for existing building is launched.
Data of 504 multi-family housing complexes are used for empirical analysis.
Potential problems of BECC are found in terms of classification and grading system.
Improved BECC is developed by the energy benchmarking and modified grading process.
Comparison between the current and improved BECC is conducted for validation.

a r t i c l e

i n f o

Article history: Received 22 September 2016 Received in revised form 7 February 2017 Accepted 12 February 2017

Keywords: Operational rating Energy policy Energy benchmark Hypothesis testing Hierarchical cluster analysis K-means clustering

a b s t r a c t The Building Energy Consumption Certification (BECC) evaluating the energy performance of existing buildings has been launched since 2016 to reduce the operational energy consumption in existing buildings in South Korea. However, the current BECC has some potential problems, and these problems should be solved to evaluate the energy performance of existing building more accurately. Thus, this study aims to identify the potential problems in the current BEEC using the hypothesis testing. And then this study proposes the improved BECC using the energy benchmarking process and the modified grading process to solve the potential problems. As a result of the hypothesis testing based on the data of 504 multi-family housing complexes (MFHCs), the potential problems were identified as follows: (i) the current classification criteria caused the irrational judgements, and (ii) the current grading system was lacking in its assessment function (over 94% of MFHCs ranked in the average level as grades ‘‘C” and ‘‘D”). To solve these problems, this study proposed the improved BECC. The energy benchmarking process provides the reasonable classification criteria, and the modified grading process finds the reasonable number of grades and its range. The result of comparative analysis based on 504 MFHCs indicated that the improved BECC could solve the problems in the current BECC. That is, over 94% of MFHCs were ranked in grades ‘‘C” and ‘‘D” in the current BECC while they were shown in all five grades (i.e., grades ‘‘A”, ‘‘B”, ‘‘C”, ‘‘D”, and ‘‘E”) in the improved BECC. The policy-makers can more accurately assess the energy performance of existing MFHCs by using the improved BECC. Ó 2017 Elsevier Ltd. All rights reserved.

1. Introduction Various political actions (e.g., regulations and supportive policies) have been implemented in the developed countries to reduce the greenhouse gas emissions and energy consumption in the building sector, which accounts for about 40% of the global primary energy consumption [1–3]. The primary energy consumption in existing multi-family housing complexes (MFHCs) accounts for ⇑ Corresponding author. E-mail address: [email protected] (T. Hong). http://dx.doi.org/10.1016/j.apenergy.2017.02.036 0306-2619/Ó 2017 Elsevier Ltd. All rights reserved.

about 35% of the total amounts of building sector in South Korea [4]. Thus, political actions (e.g., the penalty to the buildings which have relatively poor energy efficiency or the incentive to facilitate the voluntary energy savings by the residents) are required to manage the energy consumption of the existing MFHCs. The representative green building policies in South Korea can be summarized as follows: (i) Green Standard for Energy and Environmental Design (G-SEED) as green building certification; (ii) Building Energy Efficiency Rating (BEER) as energy performance certificates (EPCs) based on the simulated energy demand; and (iii) Building Energy Consumption Certification (BECC) as EPCs based on the actual

J. Jeong et al. / Applied Energy 193 (2017) 112–124

113

Nomenclature BECC BEER CEI EPBD EPCs EU EUI

Building Energy Consumption Certification Building Energy Efficiency Rating CO2 emission intensity Energy Performance Of Buildings Directive energy performance certificates European Union energy use intensity

energy consumption data [5–15]. While the G-SEED and BEER evaluate the energy performance of new buildings based on the estimated operational energy demand, the BECC evaluates the energy performance of existing buildings based on the actual operational energy consumption. In this regard, the BECC can be an effective policy to evaluate the energy performance of the existing MFHCs. Some previous studies have explored the potential problems in the current BECC [16–18]. For instance, the current BECC may lead to the irrational judgements for the energy performance of the existing MFHCs since it does not reflect the difference of climate conditions due to subdividing MFHCs based on the local government jurisdictions and most of the MFHCs could be included in the same grade. Despite these potential problems, there is a lack of studies for identifying the potential problems in the current BECC and improving the current BECC for the successful implementation of the BECC in South Korea. Therefore, this study, as a follow-up research by Jeong et al. [16], aims to identify the potential problems in the current BECC using the hypothesis testing, and to propose the improved BECC solving the potential problems of the current BECC. The improved BECC is proposed using the energy benchmarking process and the modified grading process. This study uses the energy benchmarking process developed by Jeong et al., which helps to determine accurate energy benchmarks and classify the benchmarking groups through the statistically proven data-mining techniques [16] (for the model of the energy benchmarking process, refer to Fig. S1 in the supplementary material). And the modified grading process is newly developed in this study. In addition, to verify the validity of the improved BECC, this study compares the results between the current BECC and the improved BECC based on 504 MFHCs. The improved BECC proposed in this study can more accurately evaluate the energy performance of MFHCs by solving the potential problems of the current BECC. Ultimately, it is expected to lead to a reduction in the operational energy consumption of MFHCs. Section 2 describes energy performance certificates using the operational rating and the potential problems of the current BECC via literature review. Section 3 describes the method to demonstrate the potential problems of the current BECC through statistical analysis and the improved BECC, which uses the energy benchmarking process and the modified grading process, in order to solve the potential problems of the current BECC. In Section 4, the validity of the improved BECC is demonstrated by applying 504 MFHCs to the improved BECC and the current BECC. Finally, Section 5 includes the brief description of the results and limitations of this study, and future study.

2. Review of the Building Energy Consumption Certification (BECC) SYSTEM 2.1. Energy performance certificates using the operational rating The operational rating evaluates the energy performance of existing buildings based on the actual energy consumption. Also,

G-SEED MAPE MFHC RMSE SSE WCSS

Green Standard for Energy and Environmental Design mean-absolute-percentage error multi-family housing complexes root-mean-square error sum of squared error within-cluster sum of squares

various building information on physical attributes (e.g., area, axis, floors, etc.) and socioeconomic attributes (i.e., residents, housing price, etc.,) should be considered to establish the operational rating [16,17,19–21]. That is, since the operational rating considers various building attributes that can affect the building energy consumption, it is proper to evaluate the energy performance of existing buildings. The European Union (EU) initiated the Energy Performance of Buildings Directive (EPBD) in 2002 to reduce the CO2 emissions from building sector. The EPBD defined a regulation that made it compulsory to evaluate the energy performance of new and existing buildings, and it included a clause that made it mandatory to attach the EPCs on the contract documents of buildings. Based on the EBPD guideline, the U.K., Germany, France and some EU countries established the regulation for the EPCs using the operational rating [22–26]. Similarly to the EPBD (e.g., Display Energy Certificates in U.K), the operational rating provided by the BECC in South Korea should be attached to the contract documents. The BECC has been implemented from 2013 through 2015 on a pilot basis, and it has been officially applied to over 500 households of MFHCs in 2016 [27]. While the site energy use intensity (EUI) of target building is used as an evaluation index to determine the grade of the EPCs, the source EUI and CO2 emission intensity (CEI) are presented for reference regardless of evaluation [28,29]. Though some studies has pointed the expected problems in the current BECC, they didn’t provide the alternative ways for addressing the potential problems with the scientific validations [16,18]. As shown in Table 1, the previous studies have analyzed the historical energy consumption data with the statistical value to evaluate the building energy performance as the operational rating. These studies have used various methodologies such as regression analysis, artificial neural network, data envelope analysis, multiple decision-making approach methods, and data-mining techniques. They commonly had the process to find the reliable energy benchmark with the homogeneous condition considering the attributes of buildings. In this regard, the expected problems in the current BECC can be identified through the validation of the energy benchmark in the current BECC. 2.2. Potential problems in the current BECC The clauses, which correspond to the operational rating system in the current BECC, can be summarized as follows: (i) the operational rating is evaluated by the districts based on the local government jurisdictions; (ii) the benchmarking clusters are classified by average enclosed area (AEA) based on the criteria of the Korean Census (refer to Table 2) [30]. The AEA stands for the area divided the total enclosed area by the number of households, and has been used as a unit size of households in MFHCs in South Korea [16]; (iii) the results of the operational rating in the current BECC are provided by five grades (refer to Table 3); (iv) the energy performance of the MFHCs are evaluated by site EUI including heat and electricity energy; (v) the mean value of site EUI is used as the benchmark;

114

J. Jeong et al. / Applied Energy 193 (2017) 112–124

Table 1 Literature review on the building energy benchmarking. Method

Author

Evaluation index

Building

Location

Regression analysis

Chung et al. [43] Chung and Hui [44] Xuchao et al. [45] Amini et al. [46]

Energy use intensity (EUI)

Hong Kong

EUI, CO2 emissions intensity (CEI) Electricity cost

Commercial Office Hotel Residential

Yalcintas[47] Yalcintas and Ozturk [48] Hong et al. [49] Hong et al. [50]

Weighted EUI

Office

US

EUI Fossil fuel consumption, electricity consumption

School

UK

Data envelope analysis

Lee [51] Lee and Lee [52] Lee [53]

EUI Energy consumption Cooling energy consumption

Office

Taiwan

Multiple attribute decision-making approach

Lee [54] Lee and Lin [55] Lee and Lin [56]

Floor area, number of occupants, temperature, rain hours

Office

Taiwan

Data-mining

Jeong et al. [16] Park et al. [57]

Site EUI, source EUI, CEI Source EUI

Residential Office

South Korea

Artificial neural network

Singapore US

Table 2 Benchmarking clusters by AEA in the current BECC. Criteria

Cluster one

Cluster two

Cluster three

Cluster four

Cluster five

Cluster six

AEA (m2)

Under 50

50–75

75–105

105–170

170–205

Over 205

Table 3 Grades of the operational rating for MFHCs in the current BECC. Operational rating

Grade

Note

Under 50% 50–75% 75–100%

A B C

Energy performance is better than the benchmark

100–125% Over 125%

D E

Energy performance is worse than the benchmark

and (vi) the type of heat sources (i.e., district heating and individual boiler) of MFHCs is not considered to subdivide the benchmarking group. Based on the aforementioned clauses, the problem analysis was conducted through the literature review, expert interviews, and preceding studies [16,17,28,29]. And then, the potential problems in the current BECC system were derived as follows.
(i) Problem with subdividing the benchmarking clusters by districts: The benchmarking clusters are subdivided by the districts based on the local government jurisdictions. For instance, the MFHCs located in Seoul city are subdivided into 25 administrative districts. The subdivision based on the local government jurisdictions may be convenient for the local government to manage the MFHCs. However, it can cause the irrational judgement in terms of the operational rating, because the energy benchmark can be varied by climate conditions.
(ii) Problem with clustering MFHCs by AEA: In the current BECC, the MFHCs are clustered into five groups by the AEA which is established as the clustering criteria based on Korean Census. However, in order to distinguish MFHCs more reasonably and to ensure their validity, the clustering criteria should be defined throughout the scientific analysis [16]. The current BECC without the scientific analysis can cause the irrational judgement that most MFHCs are included in certain groups.
(iii) Problem with the grading system: In case of the energy performance evaluation for new buildings based on the estimated

energy demand, the plug-in loads by appliances and computers may not be considered. Thus, it is possible to obtain the quite low level of energy demand by using various energy saving techniques reducing the heating, cooling and lighting energy demand. However, the actual energy consumption of a MFHC is used in the current BECC. For instance, in order to get the grade ‘‘A” in the current BECC, the actual energy consumption of a MFHC should be less than 50% of the average actual energy consumption of MFHCs in the relevant benchmarking cluster. Since the plug-in energy consumptions occupy a significant portion of the actual energy consumption, those of a MFHC are necessarily included in the actual energy consumption. Thus, the actual energy consumption of high-ranked MFHCs may not be significantly different compared to those of low-ranked MFHCs in the current BECC. Actually, the average site EUI for the upper 25% of 504 MFHC in this study was calculated by 96.01 kW h/ m2y and it was only 84.5% of the average site EUI for whole 504 MFHCs (e.g., 113.52 kW h/m2y).
(iv) Problem with the use of the site EUI as evaluation index: It was already proven in a preceding research by Jeong et al. [16] that the source EUI or CEI is better to measure the building energy performance of existing buildings in terms of the governmental aspect. Since this study considered site EUI, source EUI, and CEI, this problem is not considered in this study.
(v) Problem with the use of the mean value of the site EUI as the benchmark: It was also proven in the preceding research that the median value could work better as a benchmark than the mean value due to the skewness of energy usage distributions [16,31,32]. Thus, since this study considers the median value instead of the mean value, this problem is not considered in this study.
(vi) Problem with the type of heat sources: Three types of heat sources are mainly applied to the MFHCs in South Korea (i.e., LNG-based individual boiler, district heating, and central supply boiler). Since each heat source has different efficiency and primary energy source, it may be unfair that the MFHCs using different heating sources are evaluated in the same group. Thus, this study collected the data of the MFHCs using district heating only, accordingly, this problem is not considered in this study.

J. Jeong et al. / Applied Energy 193 (2017) 112–124

115

Finally, this study addressed three types of potential problems in the current BECC as follows: (i) problem with subdividing by districts; (ii) problem with clustering by AEA; and (iii) problem with the grading system.

(MAPE) and root-mean-square error (RMSE) are calculated as Eqs. (2) and (3) [33].

3. Materials and methods

rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 1 Xn RMSE ¼ ðyt  yt Þ2 t¼1 n

It should be identified whether or not the potential problems in the current BECC are statistically significant in order to propose the improved BECC. First, the energy benchmarking process is used with the hypothesis testing to demonstrate the problems with subdividing by districts and clustering by AEA. Second, the modified grading process is developed using two kinds of cluster analysis (i.e., hierarchical cluster analysis and k-means clustering), which are applied to solve the problem with the grading system. Additionally, to avoid confusing, the term ‘‘cluster” is used to explain the benchmarking cluster established by the current BECC, and the term ‘‘group” means the benchmarking group established by the energy benchmarking process in this study. As shown in Fig. 1, this study is carried out as follows: (i) hypothesis testing for the problems in terms of subdividing by districts and clustering by AEA; (ii) alternative analysis for improving the current grading system using the cluster analysis; and (iii) comparative analysis of the operational rating between the current BECC and the improved BECC (which is proposed in this study using the energy benchmarking process and the modified grading process). The dataset of 504 MFHCs (i.e., the building characteristics and energy consumption) are used to conduct the hypothesis testing, the alternative analysis, and the comparative analysis (For the detail attributes of 504 MFHCs, refer to Table S1 in the supplementary material). All the MFHCs consist of several buildings and households (which have the different AEA) and the current BECC is available for each household. Thus, it is required to collect and analyze the energy consumption data from each household for accurate analysis. However, since it was impossible to collect the data from each household due to the privacy protection regulation, this study analyzes the MFHCs including households with various AEA. 3.1. Hypothesis testing for classification issues Two kinds of the hypothesis testing are established to identify whether the potential problems are significant or not (refer to Eqs. (1)(4)). Eq. (1) shows the null-hypothesis that the energy benchmarks of the benchmarking groups with same building characteristics subdivided by districts in the current BECC are identical to each other.

H0 : li1 ¼ li2 ¼ . . . ¼ lin

ð1Þ

where Ho is the null-hypothesis, lin is the energy benchmark of district n in the benchmarking group i; and benchmarking group i is developed by the energy benchmarking process. A total of 504 MFHCs are located in same climate condition, and it was identified in the preceding research that the energy benchmark of each benchmarking group is not affected by the regional attributes (i.e., longitude and latitude) [16] (For the regional information of data, refer to Fig. S2 in the supplementary material). Thus, the energy benchmark by districts in the current BECC should not have significant difference mutually in the same benchmarking group as the null-hypothesis. To reject the null-hypothesis in Eq. (1), at least one or more significant difference of energy benchmarks should be found among the n districts in the current BECC. If the null-hypothesis is rejected, the problem with subdividing by districts is identified statistically. To identify the differences of energy benchmarks by districts, mean-absolute-percentage error

MAPE ¼

 n   1X yt  yt   n t¼1 yt 

ð2Þ

ð3Þ

ˆt is the mean of the benchmark where n is the number of districts, y of n cases; and yt is the benchmark of district t. Eq. (4) shows the null-hypothesis that the energy benchmarks of all the benchmarking groups clustered by AEA are not identical to each other, in the current BECC.

H0 : l1 – l2 – . . . – l j

ð4Þ

where Ho is the null hypothesis, and lj is the energy benchmark of Cluster j by criteria of Korean Census in the current BECC. Based on the energy benchmarking process proposed by Jeong et al. [16], the energy benchmarks of all benchmarking groups should be independent mutually to conduct a fair evaluation (For the results of energy benchmarking process for 504 MFHCs, refer to Tables S2–S4 in the supplementary material). That is, the energy benchmarks of all clusters in the current BECC should have significant differences statistically as the null-hypothesis. To reject the null-hypothesis in Eq. (4), at least one or more insignificant differences of energy benchmarks should be found among the j Clusters in the current BECC. If the null-hypothesis is rejected, the problem with clustering MFHCs by AEA is identified statistically. That is, through the hypothesis testing, it is possible to assess whether or not it is appropriate to cluster MFHCs based on AEA. 3.2. Alternative analysis for grading system As shown in Table 3, the grading system in the current BECC can be easily understood in terms of the level of grades (i.e., 50%, 75%, and 125% of benchmarks). The current grading system, however, cannot consider the actual distribution of the site EUI. To solve this issue, two kinds of approaches are available: (i) using the continuous scale instead of the grade scale; and (ii) modifying the current grading system. First, the continuous scale, which provides the numerical value of the energy performance for the target MFHC, is available as the alternative of the grade scale. For instance, the EPCs of residential buildings in Germany present the range of 0–400kW h/m2year as the continuous scale. Also, the energy benchmark and the actual site EUI of the target building are displayed at this same time as the continuous scale to identify the energy performance of the target building compared to the energy benchmark [24]. The continuous scale can prevent the irrational interpretations from the grade scale (i.e., two MFHCs having 76% and 99% of the operational rating are both grade ‘‘C” in the current BECC). However, the continuous scale cannot provide the relative level of the target building among the buildings included in the same benchmarking group in terms of the energy performance. For instance, although the target MFHC can be evaluated as 90% of the energy consumption compared to the benchmark, it cannot intuitively identify how much the target MFHC is superior to other MFHCs in the same benchmarking group. For this reason, the EPCs in Sweden are provided with the continuous scale as well as the grade scale, aiming to help people to more clearly understand it since 2014 [34]. Second, the modified grading process can be considered to make up for the weakness of the grading system in the current BECC. Two kinds of cluster analysis (i.e., hierarchical cluster analysis and k-means clustering) are applied to develop the modified grading process. It can be expected that the proper number of

116

J. Jeong et al. / Applied Energy 193 (2017) 112–124

Fig. 1. Research framework.

grades and its ranges can be determined by using the modified grading process. Thus, it can minimize the irrational interpretations from the grading system in the current BECC. In the first step of the modified grading process, it is necessary to identify the proper number of grades (clusters) based on the given database. The hierarchical cluster analysis is applied to

determine the proper number of grades. Contrary to the other cluster analysis, hierarchical cluster analysis can be applied to the situation in which the number of clusters is not given [35]. The Elbow method, as the representative method of the hierarchical cluster analysis, is applied for determining the number of clusters [36]. The proposed number of clusters can be determined at k- value.

J. Jeong et al. / Applied Energy 193 (2017) 112–124

The sum of squared error (SSE) drops significantly under k-value and reaches a plateau over k-value [37]. The SSE can be explained in Eq. (5).

SSE ¼

k X X dist ðx; ci Þ2

ð5Þ

i¼1 x2ci

where SSE is the sum of squared error; k is the number of grades (clusters); x is the set of observations; and ci is the centroid of each cluster. In the second step, the k-means clustering is conducted based on the k-grades determined in the first step. The k-means clustering aims to classify n observations into k clusters, in which each observation belongs to the cluster with the nearest mean [38]. Also, the range of each grade (cluster) can be determined through the k-means clustering. Finally, each MFHC can obtain the modified grade by considering the distribution of the energy performance in the given database. The within-cluster sum of squares (WCSS) used in the k-means clustering can be explained in Eq. (6) [39].

WCSS ¼ args min

k X X  2 dist x; li

ð6Þ

i¼1 x2si

where WSCC is the within-cluster sum of squares; k is the number of grades (clusters); x is the set of observations; si is the cluster i; and li is the centroid of each cluster. 3.3. Comparative analysis between the current and the improved BECC Based on the results from the hypothesis testing and the alternative analysis, the comparative analysis is conducted between the current and the improved BECC. Contrary to the clauses of the current BECC, the improved BECC is developed using the energy benchmarking process and the modified grading process. The energy benchmarking process is applied to solve the problems with subdividing by districts and clustering by AEA, and the modified grading process is applied to make up for the weakness of the grading system in the current BECC. 4. Results and discussion 4.1. Results of the hypothesis testing 4.1.1. Problem with subdividing MFHCs by districts in the current BECC A total of 504 MFHCs were subdivided into 36 districts in the current BECC. Before subdividing by districts, Groups 1, 2, and 3, which were developed by the energy benchmarking process, were applied to avoid the influences of clustering by AEA in the current BECC (For the detailed explanation of Groups 1, 2, and 3, refer to Table S2 in the supplementary material). It is required to conduct the hypothesis testing in terms of the problem with subdividing by districts. Table 5 shows the results of ANOVA, MAPE, and RMSE for Groups 1, 2, and 3. Though the current BECC is only evaluated by the site EUI, the source EUI and CEI are also used in this study. The MAPE and RMSE were calculated by Eqs. (2) and (3) As shown in Table 4, the energy benchmarks of 36 districts had the significant mean differences for all of the benchmarking groups and evaluation indices. It means that the energy benchmarks in the same benchmarking group can be different depending on the districts in the current BECC. Based on the energy benchmarking process, the energy benchmark in the same benchmarking group is determined to be identical regardless of regional attributes. However, the energy benchmarks by 36 districts showed the significant differences at the 0.05 level for all benchmarking groups. Thus, the null-hypothesis was rejected as explained in Section 3.1, and it was

117

identified that the problem with subdividing by districts was significant. Since a total of 36 districts were too large numbers to visualize the differences of the energy benchmarks among the districts, the identical analysis was conducted for 18 cities that had bigger area than the districts (refer to Table 5). Although the relative errors of the energy benchmarks among cities were slightly decreased than those of Table 4, the energy benchmarks by 18 cities, as shown in Table 5, also showed the significant differences by ANOVA in the same benchmarking group. It means that the analysis using 18 cities is reliable to explain the problems of the current BECC. Also, if the problems of the current BECC are identified in the 18 cities, the problems should be also identified in the 36 districts. Thus, the visualizations of the problems with subdividing by districts had been conducted using 18 cities, not 36 districts. Fig. 2 shows the distributions of the benchmarks for 18 cities in Group 1, in which the AEA of MFHCs have under 96.92 m2 using the energy benchmarking process (refer to Table S2 in the supplementary material). Some cities had the significant differences from the energy benchmark of Group 1 (For the results of Groups 2 and 3, refer to Figs. S3 and S4 in the supplementary material). Based on the preceding research, the MFHCs in the same benchmarking group should be evaluated by the only one energy benchmark [16]. And, a significant correlation between the districts (cities) and the site EUI was not founded by the latitude and the longitude as the attributes of climate condition. Thus, the significant differences in the benchmarks among the cities, classified in the current BECC, are not reasonable. Fig. 3 shows the additional effects of the classification using districts in the current BECC. Even if all the MFHCs were included in the same benchmarking group (i.e., Group 1), the rating results (i.e., the grade ‘‘A” to ‘‘E”) were significantly different among the cities. For instance, five MFHCs in City 1 have lower site EUI than the energy benchmark of Group 1, but the rating of these MFHCs were determined to be grade ‘‘D”. In contrast, two MFHCs in City 5 got grade ‘‘C” with higher site EUI than the energy benchmark of Group 1. Eventually, the problem with subdividing by districts were identified (For the additional results of Groups 2 and 3, refer to Figs. S5 and S6 in the supplementary material). 4.1.2. Problem with clustering by AEA in the current BECC A total of 504 MFHCs were classified into five benchmarking clusters by AEA in the current BECC (refer to Table 2). The MFHCs, which had under 50m2 of AEA, were not found in 504 MFHCs. Thus, Cluster-one cannot be analyzed in this study. Also, the issue with subdividing by districts were not considered to conduct the hypothesis testing in terms of the problem with clustering by AEA. Tables 6 and 7 show the results of ANOVA with post hoc analysis for the hypothesis testing. Cluster-six in this study had only one MFHC, and thus, it was excluded in the post hoc analysis due to the absence of the standard deviation. As shown in Table 6, the ANOVA result of five Clusters was significant, however, the significant result of ANOVA shows that one or more Clusters have different distribution. Thus, the post hoc analysis by Scheffé’s method was applied to conduct the multiple comparisons [40]. As shown in Table 7, the result of post hoc analysis was not significant between Cluster-two and Cluster-three. It means that the energy benchmarks of Clusters assigned in the current BECC were not independent mutually. Thus, the nullhypothesis was rejected as explained in Section 3.1, and it was identified that the problem with clustering by AEA was significant. In addition, Clusters-one, five and six consisted of zero, eleven, and one MFHCs, respectively. It was considered too small to determine the operational rating with the limited number of database (i.e., 504 MFHCs). Furthermore, each cluster should be subdivided further by districts in the current BECC. Although a total of 2740

118

J. Jeong et al. / Applied Energy 193 (2017) 112–124

Table 4 Energy Benchmarks in the current BECC based on the 36 districts. Classification

Developed benchmarks

Benchmarks of 36 districts

Relative error

ANOVA results

Min.a

Max.b

MAPEc (%)

RMSEd

F

Sig.

Site EUI (kW h/m y)

Group 1 Group 2 Group 3 Mean

123.13 114.42 101.88 –

104.59 93.61 81.105 –

147.16 129.5 119.51 –

7.54 4.77 7.27 6.53

11.37 kW h/m2y 7.59 kW h/m2y 9.54 kW h/m2y 9.50 kW h/m2y

2.760 1.843 3.859 –

0.000** 0.005** 0.000** –

Source EUI (kW h/m2y)

Group 1 Group 2 Group 3 Mean

173.17 150.55 131.13 –

136.56 119.9 106.44 –

213.06 169.97 176.16 –

6.30 5.40 8.71 6.80

15.02 kW h/m2y 10.71 kW h/m2y 15.70 kW h/m2y 13.81 kW h/m2y

1.870 1.611 2.658 –

0.022* 0.023* 0.000** –

CEI (kg CO2/m2y)

Group 1 Group 2 Group 3 Mean

28.17 24.16 20.95 –

22.01 19.15 17.07 –

34.45 27.31 28.74 –

6.15 5.55 9.03 6.91

2.41 kg CO2/m2y 1.75 kg CO2/m2y 2.63 kg CO2/m2y 2.26 kg CO2/m2y

1.670 1.575 2.508 –

0.050* 0.029* 0.001** –

2

Note: a Min. stands for the minimum value among energy benchmark of 36 districts. b Max. stands for the maximum value among energy benchmark of 36 districts. c MAPE stands for mean absolute percentage error. d RMSE stands for root mean squared error. * The mean difference between 36 districts is significant at the 0.05 level. ** The mean difference between 36 districts is significant at the 0.01 level.

Table 5 Energy Benchmarks in the current BECC based on the 18 cities. Classification

Developed benchmarks

Benchmarks of 18 cities

Relative error

ANOVA results

Min.a

Max.b

MAPEc (%)

RMSEd

F

Sig.

Site EUI (kW h/m y)

Group 1 Group 2 Group 3 Mean

123.13 114.42 101.88 –

106.02 110.34 92.77 –

147.16 129.5 119.51 –

6.60 3.56 6.27 5.48

10.84 kW h/m2y 5.64 kW h/m2y 7.93 kW h/m2y 8.14 kW h/m2y

2.635 1.801 4.033 –

0.003** 0.029* 0.000** –

Source EUI (kW h/m2y)

Group 1 Group 2 Group 3 Mean

173.17 150.55 131.13 –

154.98 144.7 120.59 –

213.06 162.45 156.39 –

5.66 3.67 5.74 5.02

14.24 kW h/m2y 6.37 kW h/m2y 10.57 kW h/m2y 10.39 kW h/m2y

2.374 1.349 2.329 –

0.007** 0.164 0.007** –

CEI (kg CO2/m2y)

Group 1 Group 2 Group 3 Mean

28.17 24.16 20.95 –

25.55 23.23 19.3 –

34.45 26.21 25.07 –

5.59 3.93 5.80 5.11

2.26 kg CO2/m2y 1.10 kg CO2/m2y 1.71 kg CO2/m2y 1.69 kg CO2/m2y

2.205 1.306 2.121 –

0.013* 0.189 0.015* –

2

Note: a Min. stands for the minimum value among energy benchmark of 18 cities. b Max. stands for the maximum value among energy benchmark of 18 cities. c MAPE stands for mean absolute percentage error. d RMSE stands for root mean squared error. * The mean difference between 18 cities is significant at the 0.05 level. ** The mean difference between 18 cities is significant at the 0.01 level.

MFHCs located in Seoul and suburban cities are included in the target range of the current BECC in 2016 [41], it is difficult to collect all things due to an omission among the attributes or privacy protection regulation. Thus, the MFHCs belonged to Cluster-one, five, and six were considered too small to determine the operational rating in practice. 4.2. Results of the alternative analysis 4.2.1. Current grading system The operational rating was calculated to identify the problems of the current grading system (refer to Table 3). Before calculating the operational rating, Groups 1, 2, and 3, which were developed by the energy benchmarking process, were applied to avoid the problems with the current BECC (i.e., subdividing by districts and clustering by AEA). Also, the mean value was applied as the energy benchmarks according to the current BECC.

As shown in Table 8, the proportions of grades ‘‘C” and ‘‘D” were determined to be 94.4%, 98.1%, and 94.3% in Group 1, Group 2, and Group 3, respectively. Also, there was no grade ‘‘A” among 504 MFHCs. It means that the MFHCs having the energy performance with a top 5% and the average level (e.g., 100% of the benchmark in Group 1) can be ranked as grade ‘‘C” identically. Furthermore, these irrational situations were found in grade ‘‘D”, which means that most of MFHCs having worse than the benchmark were ranked to ‘‘D”. The grading system can be recognized as the domain of policy maker’s strategic decision, not the scientific assessment. However, it goes against the common sense that the only 1% of MFHCs can obtain grade ‘‘B”. Also, over 94% of MFHCs ranked in grades ‘‘C” and ‘‘D” can be understood as the average level. Since residents can only find the operational rating of the target MFHCs (where they want to live or purchase), they cannot recognize how many MFHCs would be ranked in grades ‘‘C” or ‘‘D” in the current BECC.

J. Jeong et al. / Applied Energy 193 (2017) 112–124

119

Fig. 2. Distributions of the benchmarks for 18 cities in Group 1.

Fig. 3. Rating results of 18 cities in Group 1 in the current BECC.

Thus, the current BECC cannot provide the effective evaluation result for users. In this aspect, the modified grading process were proposed using the cluster analysis.

In addition, it was also proven in the preceding research that the median value is better benchmark than the mean value due to the skewness of energy usage distributions [28,31,32,42]. As

120

J. Jeong et al. / Applied Energy 193 (2017) 112–124

Table 6 ANOVA result of clusters-two, three, four, five, and six in the current BECC. Type

Cluster

No. of MFHCs

Mean

Site EUI

Two Three Four Five Six

74 176 242 11 1

121.40 118.18 108.58 95.88 81.56

95% C.I.a Min

Max

118.52 116.05 106.98 87.92 –

124.28 120.31 110.19 103.83 –

F

Sig.

26.875

0.000**

Note: a C.I. stands for confidence interval. ** The mean difference between Groups is significant at the 0.01 level.

Table 7 Post hoc analysis of Clusters two, three, four, and five in the current BECC. Type

Site EUI

M.D.a

Cluster

Two

Three Four Five Two Four Five Two Three Five Two Three Four

Three

Four

Five

3.220 12.829* 25.527* 3.220 9.600* 22.307* 12.819* 9.599* 12.708* 25.527* 22.307* 12.708*

S.E.b

1.831 1.755 4.270 1.831 1.309 4.106 1.755 1.309 4.073 4.270 4.106 4.073

Sig.

95% CI

0.378 0.000 0.000 0.378 0.000 0.000 0.000 0.000 0.022 0.000 0.000 0.022

Min.

Max.

1.915 7.896 13.550 8.354 5.928 10.788 17.742 13.271 1.282 37.503 33.826 24.134

8.354 17.742 37.503 1.915 13.271 33.826 7.896 5.928 24.134 13.550 10.788 1.282

Note: a M.D. stands for the mean difference. b S.E. stands for the standard error. * The mean difference between Groups is significant at the 0.05 level.

Table 8 Results of operational rating based on the current grades. Type

Grade

The number of cases (Proportion%) Group 1

Group 2

Group 3

Site EUI

A B C D E

Under 50% 50–75% 75–100% 100–125% Over 125%

– 2(1.9%) 55(50.9%) 47(43.5%) 4(3.7%)

– 3(1.2%) 120(46.9%) 131(51.2%) 2(0.8%)

– – 42(30%) 90(64.3%) 8(5.7%)

Sum

–

108(100%)

256(100%)

140(100%)

Note: The cases of grades ‘‘A”, ‘‘B”, and ‘‘C” mean that they consume less energy than the benchmark.

shown in Table 8, the upper and lower grades in Group 3 were not equally distributed when the mean value was applied as the energy benchmark. 4.2.2. Modified grading process As explained in Section 3.2, the modified grading process was developed using two kinds of cluster analysis. First, the hierarchical cluster analysis was conducted for a total of 504 MFHCs, and then, the Elbow method was applied to determine the proper number of grades based on the result of the hierarchical cluster analysis. Second, the k-means clustering was applied to determine the boundary value among the determined grades. Fig. 4 shows the result of the hierarchical cluster analysis and the Elbow point (where the SSE is dramatically decreased just before being stable). The Elbow point, indicating the proper numbers of grades, was determined to be five grades. The SSE was decreased under the five grades, and then, the decreasing level of the SSE was stable. Thus, the proper number of grades was

determined to be five grades, and, as a result, it was identical to the number of grades in the current BECC. Table 9 shows the result of the k-means clustering for five grades based on the hierarchical cluster analysis. The grading system in the current BECC is subdivided into the upper grade (i.e., grades ‘‘A”, ‘‘B”, and ‘‘C”) and the lower grade (i.e., grades ‘‘D” and ‘‘E”) by comparing the energy benchmark. Thus, the kmeans clustering was conducted with the premise of the current BECC. As shown in Table 10, the upper boundary of grade ‘‘A” was determined to be 78.0–84.2% compared to the mean value of the site EUI in each benchmarking group. The range of grade ‘‘B” was determined to be 78.0–93.4%; the range of grade ‘‘C”, 90.7–100%; and the range of grade ‘‘D”, 100–119%; the lower boundary of grade ‘‘E”, 109.6–119%, respectively. It was significantly different from the range of the current grading system (i.e., 50%, 75%, 100%, and 125%). However, the range of each grade was not exactly same by grades (i.e., Group 1, 2, and 3). The integration of criteria

121

J. Jeong et al. / Applied Energy 193 (2017) 112–124

Fig. 4. Result of the hierarchical cluster analysis and the Elbow point.

Table 9 Results of the k-means clustering for five grades in Groups 1, 2, and 3.

Group 1 Grade A B C D E

Group 2

Group 3

Ctda

Range

Ctda

Range

Ctda

Range

90.8 107.1 118.0 131.2 154.8

under 78.1% 78.1 to 90.7% 90.7 to 100% 100 to 117.1% over 117.1%

84.5 100.4 110.5 119.5 131.0

under 78.0% 78.0 to 92.1% 92.1 to 100% 100 to 109.6% over 109.6%

82.0 92.1 98.4 109.2 129.3

under 84.2% 84.2 to 93.4% 93.4 to 100% 100 to 119.0% over 119.0%

Note: a Ctd stands for the centroid site EUI of each grade and its unit is kW h/m2y; Grey shaded cases mean that they consume less energy than their benchmark.

Table 10 Proposed range of each grade using the modified grading process. Boundary of grade

Group 1 (A) (%)

Group 2 (B) (%)

Group 3 (C) (%)

Proposed range (A + B + C)/3 (%)

A–B B–C C–D D–E

78.1 90.7 100.0 117.1

78.0 92.1 100.0 109.6

84.2 93.4 100.0 119.0

80.1 92.2 100.0 115.2

was required to evaluate the entire MFHCs with the identical range of grades. The average ranges from three groups were proposed in this study (refer to Table 10). 4.3. Results of the comparative analysis The comparative analysis was applied to compare the operational rating between the current BECC and the improved BECC. Particularly, the energy benchmarking process and the modified grading process were applied to develop the improved BECC. The preconditions for the current BECC were as follows; (i) 504 MFHCs were classified into the six clusters by AEA and subdivided by 18 cities based on the current BECC. As a result, a total of 53 sub-clusters were created; (ii) the existing five grades were used; (iii) the mean value of each benchmarking cluster was applied as the benchmark; and (iv) the site EUI was applied as the evaluation index.

The preconditions for the improved BECC were as follows; (i) 504 MFHCs were classified into Groups 1, 2, and 3 based on the energy benchmarking process, and the regional attributes (i.e., cities or districts) were not considered; (ii) the modified five grades, established by the modified grading process, were used; (iii) the median value of each benchmarking group was applied as the benchmark; and (iv) the site EUI was applied as the evaluation index. Table 11 shows the results of the operational rating in the current BECC. As mentioned in Section 4.1.2, there was the only one MFHC in Cluster-six, and thus, it was excluded from the comparative analysis. Thus, even if the current BECC provides six clusters by AEA, three clusters with enough number of MFHCs were considered in this study. Due to the subdivision by 18 cities, 4.54–6.21% of MAPE were found in the same cluster, and such relative errors can cause the irrational judgements for the MFHCs having similar energy performance. Even though the upper (i.e.,

122

J. Jeong et al. / Applied Energy 193 (2017) 112–124

Table 11 Results of the operational rating in the current BECC. Grade

Cluster two (AEA: 50–75 m2)

Cluster three (AEA: 75–105 m2)

Cluster four (AEA: 110–170 m2)

Cluster five (AEA: 170–205 m2)

No. of MFHCs Benchmark (kW h/m2y) Sub-clusters by districts (cities) MAPE RMSE (kW h/m2y)

74 121.40 14 5.59% 9.98

176 118.18 18 4.54% 7.11

242 108.58 17 6.21% 8.24

11 95.88 3 4.99% 4.83

Current BECC

A B C D E

Under 50% 50–75% 75–100% 100–125% Over 125%

Sum

0 39 35 0

(0.0%) (0.0%) (52.7%) (47.3%) (0.0%)

0 4 95 70 7

(0.0%) (2.3%) (54.0%) (39.8%) (4.0%)

0 0 134 104 4

(0.0%) (0.0%) (55.4%) (43.0%) (1.7%)

0 0 6 5 0

(0.0%) (0.0%) (54.5%) (45.5%) (0.0%)

74

(100%)

176

(100%)

242

(100%)

11

(100%)

grades ‘‘A”, ‘‘B”, and ‘‘C”) and the lower grades (i.e., grades ‘‘D” and ‘‘E”) were divided by the energy benchmark, the ratio of the upper and the lower grades were not equally distributed in the current BECC. In addition, 93.8–100% of MFHCs were ranked in grades ‘‘C” or ‘‘D” among clusters. As a result, the current BECC caused the various problems in calculating the operational rating of the MFHCs.

Table 12 shows the results of the operational rating in the improved BECC. The Groups 1, 2, and 3 by the energy benchmarking process had enough number of MFHCs, respectively, and the mean differences of the benchmarks had been validated as significant using ANOVA with post hoc analysis. When applying the modified grading process, the upper and the lower grades for Groups 1, 2, and 3 were provided equally in the improved BECC.

Table 12 Results of the operational rating in the improved BECC. Grade

Group 1 (AEA: under 96.92 m2)

Group 2 (AEA: 96.92– 135.46 m2)

Group 3 (AEA: over 135.46 m2)

No. of MFHCs Benchmark (kW h/m2y)

108 123.13

256 114.42

140 101.88

Improved BECC

Sum

A B C D E

Under 80.1% 80.1–92.2% 92.2–100% 100–115.2% Over 115.2%

3 26 25 40 14

(2.8%) (24.1%) (23.1%) (37.0%) (13.0%)

4 41 83 113 15

(1.6%) (16.0%) (32.4%) (44.1%) (5.9%)

5 20 45 59 11

(3.6%) (14.3%) (32.1%) (42.1%) (7.9%)

108

(100%)

256

(100%)

140

(100%)

Fig. 5. Comparison of the operational rating between the current and the improved BECC.

J. Jeong et al. / Applied Energy 193 (2017) 112–124

In addition, 17.6–26.9% of 504 MFHCs were ranked in grades ‘‘A” and ‘‘B”, and 5.9–13% of 504 MFHCs were ranked in grade ‘‘E”. Fig. 5 shows the comparison of the operational rating between the current BECC and the improved BECC. While most of the MFHCs were ranked in grades ‘‘C” and ‘‘D” in the current BECC throughout every clusters, they were shown in all of five grades (i.e., grades ‘‘A”, ‘‘B”, ‘‘C”, ‘‘D”, and ‘‘E”) in the improved BECC. For instance, 52.7% and 47.3% of MFHCs in the cluster two are included in grades ‘‘B” and ‘‘C” in the current BECC, respectively. However, there is no MFHCs included in grades ‘‘A”, ‘‘D”, and ‘‘E”. On the other hand, in the improved BECC, MFHCs are distributed in all five grades although there are relatively few MFHCs included in grade ‘‘A” (e.g., 2.8% of MFHCs are included in grade ‘‘A” in Group 1 the improved BECC). Therefore, it was determined that the energy performance of 504 MFHCs can be evaluated more clearly in the improved BECC. 4.4. Discussion The current BECC has classified the benchmarking clusters by considering the districts based on the local government jurisdictions in South Korea as well as the AEA based on the criteria of the Korean Census. It can be reasonable in terms of the administrative management. The operational rating based on the actual energy consumption, however, should be determined by considering the building characteristics. From the perspective of the energy benchmarking, the problems of the current BECC were identified statistically in this study. It is not reasonable to assign the different grades for two identical MFHCs located in other districts. For improving the reliability of the current BECC, the problematic clauses should be revised. Generally, the grading system for determining the number of grades and its range could be understood as the domain of policy maker’s strategic decision, not the scientific assessment. This study, however, found the irrational results of the current BECC (i.e., over 94% of MFHCs ranked in the average level as grades ‘‘C” and ‘‘D”). In this regard, the modified grading process was proposed to provide the reasonable number of grades and its range. The proper number of grades can be determined using the hierarchical cluster analysis with the Elbow method. Also, the k-mean clustering can provide the reliable range of each grade by considering the statistical distance of each case. The modified grading process has an advantage in terms of presenting the scientific answers to the subjective issues. And, it can be widely used to determine the grades of various database, of which the detailed criteria are not established. 5. Conclusions The BECC is an effective policy to evaluate the energy performance of existing MFHCs. However, some potential problems were found in the clauses of the current BECC. Therefore, this study aimed to identify the potential problems in the current BECC and to propose the alternatives for improving the BECC. First, by conducting the hypothesis testing based on the data of 504 MFHCs, the potential problems in the current BECC were identified. As a result, three kinds of potential problems in the current BECC were mainly analyzed as follows: (i) the problem with subdividing the benchmarking clusters by districts based on the local government jurisdictions; (ii) the problem with clustering MFHCs by AEA based on the criteria of the Korean Census; and (iii) the problem with the grading system where most MFHCs are biased to be included in the same grade. Second, in order to solve these potential problems causing the irrational judgements and improve the current BECC, the improved BECC consists of the energy bench-

123

marking process and the modified grading process. The energy benchmarking process solves first and second potential problems by clustering MFHCs with the consideration of AEA. The modified grading process solves a third potential problem by determining the proper number of grades and its ranges based on the distribution of actual energy consumption. To verify the validity of the improved BECC, 504 MFHCs were applied to the current BECC and the improved BECC, and the results were compared. As a result, MFHCs were biased in grades ‘‘C” and ‘‘D” in the current BECC (over 94% of MFHCs were ranked in grades ‘‘C” and ‘‘D”), while they were shown in all five grades (i.e., grades ‘‘A”, ‘‘B”, ‘‘C”, ‘‘D”, and ‘‘E”) in the improved BECC. For instance, 108 MFHCs included in Group 1 were distributed into five grades (i.e., grade ‘‘A” 2.8%; ‘‘B” 24.1%; ‘‘C” 23.1%, ‘‘D” 37.0%; and ‘‘E” 13.0%). That is, the problematic situation, which over 94% of MFHCs were ranked in grades ‘‘C” and ‘‘D”, was improved by using the improved BECC. It indicates that the improved BECC could solve the problems in the current BECC. Accordingly, the improved BECC can be considered to be reasonable and more accurately evaluate the energy performance of existing MFHCs. This study proposed the improved BECC and verified the validity of the improved BECC based on 504 MFHCs. However, since the results can vary depending on the data used, the additional data of MFHCs should be collected and analyzed in order to obtain the results that can be used to actually amend the current BECC. In addition, the criteria for establishing the operational rating of the building may vary depending on the types of buildings. In the future study, therefore, it is necessary to verify the validity of the operational rating system for other types of buildings as well as the MFHC. Toward this end, the future study should collect additional data on other types of building such as office and educational buildings and apply the proposed method to the additionally collected data. Acknowledgement This work was supported by the National Research Foundation of Korea (NRF) grant funded by the Korea government (MSIP; Ministry of Science, ICT & Future Planning) (NRF2015R1A2A1A05001657). Appendix A. Supplementary material Supplementary data associated with this article can be found, in the online version, at http://dx.doi.org/10.1016/j.apenergy.2017. 02.036. References [1] Hong T, Koo C, Kim J, Lee M, Jeong K. A review on sustainable construction management strategies for monitoring, diagnosing, and retrofitting the building’s dynamic energy performance: focused on the operation and maintenance phase. Appl Energy 2015;155:671–707. [2] United Nations Framework Convention on Climate Change (UNFCCC). Kyoto protocol to the United Nations Framework Convention on Climate Change. Kyoto, United Nations (UN); 1998. [3] Intergovernmental Panel on Climate Change (IPCC). Climate change 2007. Synthesis report. Geneva: IPCC; 2007. [4] Ministry of Trade, Industry and Energy (MOTIE). 2012 Primary energy consumption by building type. MOTIE; 2013. [5] Korea Energy Agency (KEA). Certification information for building energy efficiency rating. KEA; 2015. [6] Korea Environmental Industry and Technology Institute (KEITI). G-SEED certified projects. KEITI; 2015. [7] Korea Energy Agency (KEA). Building energy efficiency rating system. KEA; 2015. [8] Chun J, Son W, Shin J, Park K. Evaluation of building energy efficiency rating of the business buildings: a case study. The Society Of Air-Conditioning And Refrigerating Engineers Of Korea; 2011. p. 673–6. [9] Seoul city. Seoul green building design criteria. Seoul City: Housing Department; 2013.

124

J. Jeong et al. / Applied Energy 193 (2017) 112–124

[10] Seoul city. Seoul green design guideline. Seoul City: Housing Department; 2010. [11] Ministry of Land, Infrastructure and Transport (MOLIT). Operational regulation for Building Energy Efficiency Rating system. MOLIT; 2011. [12] Ministry of Land, Infrastructure and Transport (MOLIT). Regulation of heat transmission coefficient for residential buildings. MOLIT; 2010. [13] Ministry of Land, Infrastructure and Transport (MOLIT). G-SEED Certification Criteria. MOILT; 2010. [14] Mok S, Cho D. Comparison criteria and certified scores between G-SEED and LEED certification systems in office building cases. World SB14 Barcelona; 2014. [15] Wang J, Kim J, Lee K, Park I. G-SEED: the revised Korean green building certification system. In: 30th International PLEA conference. Ahmadabad: CEPT University; 2014. [16] Jeong J, Hong T, Ji C, Kim J, Lee M, Jeong K. Development of an integrated energy benchmark for a multi-family housing complex using district heating. Appl Energy 2016;179:1048–61. [17] Hong T, Koo C, Kim H, Seon Park H. Decision support model for establishing the optimal energy retrofit strategy for existing multi-family housing complexes. Energy Policy 2014;66:157–69. [18] Yang S. Improvement of building energy performance disclosure system based on energy consumption in apartment buildings. Ajou University; 2014. [19] Jeong J, Hong T, Ji C, Leigh S. Establishing a energy utilization database for energy performance evaluation of multi-family housing using district heating. Korea Inst. Constr. Eng. Manage. 2015;16:105–13. [20] Koo C, Park S, Hong T, Park HS. An estimation model for the heating and cooling demand of a residential building with a different envelope design using the finite element method. Appl Energy 2014;115:205–15. [21] Koo C, Hong T. A dynamic energy performance curve for evaluating the historical trends in the energy performance of existing buildings using a simplified case-based reasoning approach. Energy Build 2015;92:338–50. [22] Concerted Action Energy Performance of Buildings Directive (CA EPBD). Implementation of the EPBD in England and Wales. CA EPBD: Scotland and Northern Island; 2011. [23] Concerted Action Energy Performance of Buildings Directive (CA EPBD). Implementation of the EPBD in France. CA EPBD; 2011. [24] Concerted Action Energy Performance Of Buildings Directive (CA EPBD). Implementation of the EPBD in Germany. CA EPBD; 2011. [25] Department for Community and Local Government (DCLG). Improving the energy efficiency of our buildings: a guide to display energy certificates and advisory reports for public buildings. London: DCLG; 2012. [26] Department for Community and Local Government (DCLG). Display energy certificate: how efficiently is this building being used? London: DCLG; 2012. [27] Ministry of Land, Infrastructure and Transport (MOLIT). Law of supporting green building. MOLIT, editor. Korea: Ministry of Government Legislation; 2014. [28] Koo C, Hong T, Lee M, Seon Park H. Development of a new energy efficiency rating system for existing residential buildings. Energy Policy 2014;68:218–31. [29] Koo C, Hong T. Development of a dynamic operational rating system in energy performance certificates for existing buildings: geostatistical approach and data-mining technique. Appl Energy 2015;154:254–70. [30] Statistics Korea. Korea: Korean Census Statistics; 2010. [31] Jeong J, Hong T, Ji C, Kim J, Lee M, Jeong K. Development of an evaluation process for green and non-green buildings focused on energy performance of G-SEED and LEED. Build Environ 2016;105:172–84. [32] Jeong J, Ji C, Hong T, Park H. Model for evaluating the financial viability of the BOT project for highway service areas in South Korea. J Manage Eng 2016;32:04015036. [33] Chai T, Draxler RR. Root mean square error (RMSE) or mean absolute error (MAE)?–Arguments against avoiding RMSE in the literature. Geosci Model Dev 2014;7:1247–50.

[34] Adjei A. A study of homeowners’energy efficiency improvements and the impact of the energy performance certificate. In: EPBD I, editor. EPBD; 2011. [35] Mooi E, Sarstedt M. Cluster analysis. A concise guide to market research: the process, data, and methods using IBM SPSS Statistics. Berlin, Heidelberg, Springer, Berlin Heidelberg; 2011. p. 237–84. [36] Ng A. K-means algorithm in machine learning. Stanford University; 2015. [37] Kodinariya T, Makwana P. Review on determining number of cluster in Kmeans clustering. Int J Adv Res Comput Sci Manage Stud 2013;1:6. [38] Ketchen D, Shook C. The application of cluster analysis in strategic management research: an analysis and critique. Strateg Manage J 1996;17:8. [39] MacQueen J. Some methods for classification and analysis of multivariate observations. In: Proceedings of the fifth Berkeley symposium on mathematical statistics and probability, vol. 1: statistics. Berkeley (CA): University of California Press; 1967. p. 281–97. [40] Lee h, Yim J. Manual for Statistical Package for the Social Sciences. Seoul: Jiphyunjae; 2011. [41] Korea Appraisal Board. List of MFHCs over 500 households in Seoul and suburban cities. Korea; 2015. [42] Kontokosta CE. Local law 84 energy benchmarking data. New York University; 2012. [43] Chung W, Hui YV, Lam YM. Benchmarking the energy efficiency of commercial buildings. Appl Energy 2006;83:1–14. [44] Chung W, Hui YV. A study of energy efficiency of private office buildings in Hong Kong. Energy Build 2009;41:696–701. [45] Xuchao W, Priyadarsini R, Siew Eang L. Benchmarking energy use and greenhouse gas emissions in Singapore’s hotel industry. Energy Policy 2010;38:4520–7. [46] Amini M, Frye J, Ilic´ M, Karabasoglu O. Smart residential energy scheduling utilizing two stage mixed integer linear programming. In: 2015 North American power symposium (NAPS), Charlotte, NC; 2015. p. 1–6. [47] Yalcintas M. An energy benchmarking model based on artificial neural network method with a case example for tropical climates. Int J Energy Res 2006;30:1158–74. [48] Yalcintas M, Aytun Ozturk U. An energy benchmarking model based on artificial neural network method utilizing US Commercial Buildings Energy Consumption Survey (CBECS) database. Int J Energy Res 2007;31:412–21. [49] 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 Sust Built Environ. 2013;2:119–30. [50] Hong S-M, Paterson G, Mumovic D, Steadman P. Improved benchmarking comparability for energy consumption in schools. Build Res Inf 2014;42:47–61. [51] Lee W-S. Benchmarking the energy efficiency of government buildings with data envelopment analysis. Energy Build 2008;40:891–5. [52] Lee W-S, Lee K-P. Benchmarking the performance of building energy management using data envelopment analysis. Appl Therm Eng 2009;29:3269–73. [53] Lee W-S. Benchmarking the energy performance for cooling purposes in buildings using a novel index-total performance of energy for cooling purposes. Energy 2010;35:50–4. [54] Lee W-S. Evaluating and ranking energy performance of office buildings using fuzzy measure and fuzzy integral. Energy Convers Manage 2010;51:197–203. [55] Lee W-S, Lin Y-C. Evaluating and ranking energy performance of office buildings using Grey relational analysis. Energy 2011;36:2551–6. [56] Lee W-S, Lin L-C. Evaluating and ranking the energy performance of office building using technique for order preference by similarity to ideal solution. Appl Therm Eng 2011;31:3521–5. [57] Park HS, Lee M, Kang H, Hong T, Jeong J. Development of a new energy benchmark for improving the operational rating system of office buildings using various data-mining techniques. Appl Energy 2016;173:225–37.