Download as pdf or txt
Download as pdf or txt
You are on page 1of 15

Renewable and Sustainable Energy Reviews 127 (2020) 109885

Contents lists available at ScienceDirect

Renewable and Sustainable Energy Reviews


journal homepage: http://www.elsevier.com/locate/rser

A novel operation approach for the energy efficiency improvement of the


HVAC system in office spaces through real-time big data analytics
Wenzhuo Li a, Choongwan Koo b, *, Taehoon Hong c, Jeongyoon Oh b, Seung Hyun Cha d,
Shengwei Wang a
a
Department of Building Services Engineering, The Hong Kong Polytechnic University, Hung Hom, Kowloon, Hong Kong
b
Division of Architecture & Urban Design, Incheon National University, Incheon, Republic of Korea
c
Department of Architecture and Architectural Engineering, Yonsei University, Seoul, 03722, Republic of Korea
d
Department of Interior Architecture Design, Hanyang University, Seoul, 04763, Republic of Korea

A R T I C L E I N F O A B S T R A C T

Keywords: Since a traditional centralized control system (e.g., building energy management system) with a fixed schedule
Real-time big data analytics and manual control is not appropriate to irregularly occupied rooms, it is expected to have a large amount of
Energy efficiency energy saving potential in operating the HVAC system. To overcome this challenge, this study aimed to develop a
Set-point temperature
novel operation approach for the energy efficiency improvement of the HVAC system in office spaces. The real-
Change point analysis
Occupancy-based control
time indoor environmental indicators were collected and analyzed to evaluate the current operation status of the
CO2 concentration HVAC system as well as to propose a novel control strategy in two ways. The significant findings can be illus­
trated as follows. First, it could be stated that occupants would tend to establish a lower set-point temperature for
a cooler indoor environment. To solve this issue, a basic control strategy was proposed to detect the anomaly
detection of the HVAC system and to automatically adjust the indoor temperature within a preferred range.
Second, it could be evaluated that the HVAC system would be kept operating since occupants would forget to
turn off the HVAC system after the meetings. To solve this issue, an advanced control strategy was proposed to
operate the automatic on/off control of the HVAC system by considering the indoor temperature and CO2
concentration in real time. The proposed strategies can contribute to a large amount of energy savings in
operating the HVAC system of irregularly occupied spaces.

energy management system (BEMS) [4], which regulates the operation


schedule (e.g., on/off condition), the set-point temperature and volume,
1. Introduction
regardless of the different characteristics of office spaces. However, it is
required to control the HVAC system depending on the different types of
As a response to global warming, the Chinese government set the
spaces (e.g., irregular occupied offices). Reportedly, it is estimated to
2030 goal to decrease the CO2 emissions per gross domestic product by
achieve the energy savings by over 20% by carefully establishing the
60–65% from 2005 and the energy consumption per gross domestic
set-points and the operation schedules of the HVAC system [5]. There­
product by 15% from 2005 [1]. It is reported that 20–40% of the na­
fore, this study aimed to improve the energy efficiency of the HVAC
tional energy consumption come from the building sector, and especially
system in office spaces by deploying the optimal operation strategies.
for Hong Kong, the proportion is over 60% in which a commercial sector
The low-cost sensors and cloud servers have become more accessible
accounts for more than half [2,3]. The heating, ventilating and
due to the rapid growth in the Internet-of-Things (IoT) industry [6,7].
air-conditioning (HVAC) system is widely operated in buildings to keep
Accordingly, the real-time data opens new opportunities with the
a comfortable indoor environment, and its energy consumption is esti­
valuable information hidden behind, and it can be effectively used with
mated to take up the largest proportion (i.e. 25%) of the total energy
the advanced computing algorithms and technologies. In this context,
usages [3]. Thus, there is a large amount of energy saving potential in
there are a growing number of studies on the energy efficiency
operating the HVAC system of office buildings.
improvement in buildings through real-time big data analytics, such as
The HVAC system is generally controlled by the centralized building

* Corresponding author.
E-mail addresses: wendy.wz.li@connect.polyu.hk (W. Li), cwkoo@inu.ac.kr (C. Koo), hong7@yonsei.ac.kr (T. Hong), jyoh1500@gmail.com (J. Oh), chash@
hanyang.ac.kr (S.H. Cha), shengwei.wang@polyu.edu.hk (S. Wang).

https://doi.org/10.1016/j.rser.2020.109885
Received 8 March 2020; Received in revised form 29 March 2020; Accepted 25 April 2020
Available online 7 May 2020
1364-0321/© 2020 Elsevier Ltd. All rights reserved.
W. Li et al. Renewable and Sustainable Energy Reviews 127 (2020) 109885

However, these methods could cause the high costs and privacy issues.
Nomenclature Therefore, alternative ways that are cheaper and unconcerned of privacy
issues should be considered to develop the occupancy-based control
ADM accumulated duration of meeting strategies for the HVAC system.
AEURH accumulated effective utilization ratio of the HVAC To address the aforementioned challenges, a novel operation
system approach was developed to improve the energy efficiency of the HVAC
AODH accumulated operation duration of the HVAC system system in office spaces through real-time big data analytics. First, the
BEMS building energy management system set-point temperature of the HVAC system could be estimated by
CPA change point analysis calculating the mean value of indoor temperature. One of the change
DM duration of meeting point analysis (CPA) models was chosen to identify changes in the mean
EURH effective utilization ratio of the HVAC system value of indoor temperature and to evaluate the current status of the set-
HVAC system heating, ventilating and air-conditioning system point temperature in operating the HVAC system. The proposed model
IEQ indoor environmental quality can detect the potential equipment failure in the HVAC system by esti­
ODH Operation duration of the HVAC system mating the set-point temperature in real time, and it can provide feed­
VOCs volatile organic compounds backs to adjust the set-point temperature within a preferred range. On
the other hand, the indoor environmental indicators were adopted to
evaluate whether the HVAC system would be effectively operated and
whether the office spaces would be occupied. Accordingly, it is expected
pattern recognition [8–13], anomaly detection [14–17], model predic­ to effectively operate the HVAC system by applying the real-time oc­
tive control [18–21]. Tables 1 and 2 summarize previous studies on the cupancy-based control strategy.
real-time energy efficiency improvement by adjusting the set-point
temperature of the HVAC system and by deploying the 2. Materials and methods
occupancy-based control of the HVAC system.
First, the set-point temperature of the of HVAC system is generally 2.1. Research framework
used to maintain the indoor temperature within a preferred range. As
shown in Table 1, many previous studies have proved that the significant The previous studies, as a dynamic management approach in build­
energy-saving potential could be achieved by appropriately establishing ings, were conducted by the author [36,37] to achieve the improvement
the set-point temperature of the HVAC system. Ghahramani et al. [22] of the indoor environmental quality (IEQ). As shown in Supplementary
analyzed the impact of the set-point temperature, dead band, and oc­ Material (SM), Fig. S1, the dynamic management approach was adopted
cupancy schedule on energy-saving potential via energy simulation. to achieve the energy efficiency improvement in office spaces by esti­
Compared to the annual set-point temperature (i.e. 22.5 � C), the daily mating the set-point temperature and by improving the effective utili­
set-point temperature (i.e. 19.5–25.5 � C) can save energy by 6.78%– zation of the HVAC system in real time. This study established the
37.03% depending on different climates and building size. Hoyt et al. research framework into three phases.
[23] suggested to increase the cooling set-point temperature from 22.2

C to 25 � C so that the energy savings can be achieved by 29%. Both of (i) Monitoring phase (refer to section 2.2): The IoT-based sensors were
simulations and field studies supported that adequate set-point tem­ installed to detect the indoor environmental quality in real time,
perature was beneficial in terms of energy savings. in which temperature and CO2 concentration were considered.
Second, the discrepancies between the designed and actual energy (ii) Diagnostic phase (refer to section 2.3): Based on the collected
usage can mainly occur due to the occupants’ behaviors. Significant dataset, two types of diagnoses were conducted. First, the set-point
energy saving potential could be achieved if building services would temperature of the HVAC system was estimated to evaluate
effectively respond to occupancy condition. As shown in Table 2, many whether it has been maintained within a preferred range (for energy
previous studies have made efforts in developing the occupancy-based saving and thermal comfort). Second, the effective utilization of the
control strategies. Labeodan et al. [24] installed the wireless sensors HVAC system was evaluated by investigating whether the HVAC
and actuators to facilitate the occupancy-based lighting control, in system was turned off after the meeting schedule, in which temper­
which passive infrared motion sensors and chair sensors were used to ature and CO2 concentration were considered.
determine the occupants’ presence or absence. Wang et al. [25] estab­ (iii) Intervention phase (refer to section 2.4): Based on the diagnostic
lished the occupancy-based ventilation strategy for multi-zone spaces results, two types of intervention strategies were suggested. First, the
via Wi-Fi network. Compared to the traditional method with a fixed set-point temperature of the HVAC system could be estimated to
outdoor air ratio, the proposed strategy determined the proper outdoor detect the possible equipment failure and to provide proper sug­
air ratio depending on the detected and predicted occupancy level. gestions for energy saving and thermal comfort. Second, as an
Although several studies have been conducted for the energy effi­ operation strategy, the shutdown of the HVAC system was suggested
ciency improvement by investigating the set-point temperature and to avoid the unnecessary operation of the HVAC system when office
occupancy-based control in real time, there were two kinds of challenges spaces were not occupied; and thus, the effective utilization of the
to be overcome. On one hand, previous studies tried to achieve the en­ HVAC system could be achieved.
ergy savings by developing the thermostat of the HVAC system, based on
the assumption that the actual indoor temperature would reach the 2.2. Monitoring phase
designated set-point temperature. However, it could be acceptable only
if the HVAC system is operated in normal condition. That is, considering The living-lab approach was adopted to collect the indoor environ­
the accidental failures of the HVAC system (e.g. biases of temperature mental quality in office spaces, in which the qualified environmental
sensor inside, control signal errors, etc.), the designated set-point tem­ sensors were used. As a spatial-temporal scope, this study targeted two
perature of the HVAC system would not be met under the stable con­ office spaces (which were mainly used for a cooperative work in an
dition. Therefore, it is needed to accurately estimate the set-point irregular manner) in “A" office located in Hong Kong, and collected the
temperature with a concept of dead band instead of utilizing the value real-time dataset in a summer season (from Jul. 31, 2017 to Sep. 30,
shown in a control panel. On the other hand, previous studies tried to 2017). .
achieve the energy savings with the occupancy-based control strategies A centralized HVAC system was installed in “A" office, and it could be
by using passive infrared sensors, cameras, Wi-Fi signals, and so on. adjusted by occupants via a control panel. Accordingly, the set-point

2
W. Li et al. Renewable and Sustainable Energy Reviews 127 (2020) 109885

Table 1
Previous studies on the energy efficiency improvement by adjusting the set-point temperature of the HVAC system.
Author Area Building Time period Objects Tools Method Main findings

Ghahramani The Office One year The influence of EnergyPlus, Building energy � This paper introduced a
et al. (2016) United building size, MATLAB simulations, N-way ANOVA systematic approach to
[22] States construction category, software analyses study the effects of
climate, occupancy influential factors on
schedule, set point, and building HVAC energy
dead band consumption by using
building energy
simulations.
Hoyt et al. The Office From The influence of cooling EnergyPlus Simulation and empirical � It is estimated that with the
(2015) [23] United November 2010 set-point, heating set- version 7.2 and corroboration increased cooling set-point,
States to August 2012 point, and minimum the software the average cooling energy
volume flow rates of JEPlus and total HVAC energy
variable air volume on could be saved by 29% and
energy-saving of the 27% respectively.
HVAC system
� Wider set-point range can
save energy consumption
and satisfy comfort demand
by personal controls.
Wang et al. The Office January 02, Temperature set-points, Portfolio Monitoring-based � An actual energy saving of
(2013) [26] United 2010 to 02/28/ cooling coil stages, Manager, commissioning 10% was
States 2011 and damper operation, and EnergyIQ, realized through the process
January 09, optimal start EnergyPlus of ongoing commissioning.
2011 to 02/28/
2012
Yamtraipat Thailand Office N/A Adjust room N/A Field investigation � The overall electricity
et al. (2006) temperature set-point consumption saving would
[27] be 804.60 GWh/year,
which would reduce the
CO2 by 579.31 � 103 tons/
year.
Ghahramani The Office From April 1 to The influential factors N/A The human interaction for � The proposed method
et al. (2014) United June 20 and are indoor temperature, thermal comfort, fuzzy saved the energy
[28] States from October 1 set-point temperature, pattern recognition consumption and
to October 25 in airflow, outdoor algorithm, Spearman’s and maintained the thermal
2013 temperature, and Kendall’s rank correlation comfort at the same time.
subjective thermal coefficient, regression
comfort vote analysis, and optimization
Kim et al. The Office, One week in Anomaly detection for N/A Low pass filter, Z-transform, � The proposed method was
(2017) [29] United commercial different set-points and schedules peak detection algorithm, based on an optimization
States building and seasons of rooftop units (RTUs) and Bayesian classifier problem with a single
factory objective function (i.e. set-
point temperature).
� The set-point detection and
HVAC cycling algorithms
were developed based on
the peak detection tech­
nique, and they could
detect the set-points and
short cycling of RTUs
accurately.
Moon and Han The Residential January 3 and The influence of setback eQUEST Computer simulation � There was large energy-
(2011) [30] United 28, June 27 and period; set-point saving potential in both
States July 8 temperature; setback cold and hot-humid climate
temperature on energy- zones by different thermo­
saving of the HVAC stat strategies.
system

temperature and on/off condition of the HVAC system could be in this study can be defined as a “Big Data”. The five kinds of indicators
controlled in two ways. On one hand, a fixed operation schedule could in two office spaces were investigated in a nearly real-time resolution (i.
be applied to the office spaces via the BEMS with the on/off condition of e. every 1 min). That is, it exactly corresponds to “High Volume”, “High
the HVAC system, indicating that it automatically turns on at 7:00am Velocity”, and “High Variety”. It should be noted that temperature and
and turns off at 23:00pm. On the other hand, the HVAC system could be CO2 concentration were mainly investigated to develop a novel opera­
turned on/off by occupants and the relevant set-point temperature could tion approach for the energy efficiency improvement of the HVAC sys­
be adjusted by their preference. tem in office spaces. Supplementary Material (SM), SM Table S1 shows
The environmental sensors (AWAIR) can collect the environmental the details of the specifications of the environmental sensor.
indicators, such as temperature, humidity, CO2 concentration in real
time. The sensor collects the environmental data in 1-min interval; 2.3. Diagnostic phase
however, the data are automatically calculated to the 15-min averaged
value in a cloud server. The previous study conducted by De Mauro et al. Since the HVAC system is generally installed to maintain a
[38] defined a “Big Data” as a dataset characterized by high volume, comfortable indoor environment in “A" office, the real-time environ­
high velocity and high variety. With the definition, the dataset collected mental indicators were used to investigate the operation condition of the

3
W. Li et al. Renewable and Sustainable Energy Reviews 127 (2020) 109885

Table 2
Previous studies on the energy efficiency improvement by deploying the occupancy-based control of the HVAC system.
Author Area Building Time period Objects Time Tools Method Main findings
scale

Labeodan et al. Netherlands Office Three weeks Energy saving for N/A Passive Experimental study � Energy was saved by means
(2016) [24] lighting system infrared of occupancy-driven lighting
motion control.
sensors
Wang et al. Hong Kong Multi- weekday, CO2 concentration 1 min TRNSYS, Simulation � Occupancy pattern was
(2018) [25] zone weekend, and and energy WiFi devices, analyzed to determine a
office holiday consumption and camera proper outdoor air
circulation level, resulting in
energy savings while
sacrificing the indoor
environmental quality,
which performs better than
the traditional one.
Manjarres et al. Spain Office 63 days in Save energy by 1h Python The random forest � The HVAC ON/OFF and
(2017) [31] summer and optimal schedule (RF) regression mechanical ventilation (MV)
46 days in operation of the HVAC techniques operation schedule could be
winter ON/OFF and predictively controlled
mechanical based on 24-h ahead pre­
ventilation diction of indoor tempera­
ture by RF regression
techniques.
Moon and Kim South Korea Hotel From June 1st Reduce the cooling 1 min TRNSYS and ANN and simulation � The cooling energy
(2016) [32] to September energy consumption MATLAB consumption required to
30th in 2014 and improve thermal software maintain the indoor
comfort temperature within a
comfortable level was
estimated by ANN model,
which considered a normal
set-point temperature dur­
ing unoccupied period.
Yang and The United Office From January Save energy by 3 min OpenStudio ARMA, NN, Markov � The occupancy-profile-
Becerik- States to April in operating HVAC chain, logit regression, based HVAC schedules could
Gerber (2014) 2013 according to bootstrapping- save 9% of energy.
[33] occupancy profile, derivative method, k- � Room reassignment was
and by rearranging means clustering, and achieved by “P rule”, “S
occupants to different simulation rule” and “C rule” in
zones clustering process.
� Further 8% energy was
saved by coupling
occupancy-profile-based
control schedule with room
reassignment.
Nagarathinam Indian Open- From 1 Thermal comfort and 1 and N/A Temperature and � Considering the spatial
et al. (2017) plan October in energy consumption 15 humidity prediction variation of mean radiant
[34] office 2015 to 31 min EnergyPlus Simulation temperature and occupancy,
January in and MLEþ a model based predictive
2016 MATLAB Optimization control was developed to
regulate the dynamic
temperature set-point via
the PID controller.
Nikdel et al. The United Office N/A The efficiency of N/A EnergyPlus Computer simulation � The occupancy-based con­
(2018) [35] States occupancy-based trol performed effectively in
HVAC controls in energy cost, fossil fuels con­
energy cost, fossil sumption and gas emissions,
fuels consumption and and the performance
gas emissions improvement varied across
different climate zones and
HVAC systems.

HVAC system and the potential of the energy efficiency improvement of system via the control panel. Regulations or standards in different cli­
the HVAC system. Considering the complexity of analysis methods (e.g. mates were established for occupants to determine the set-point tem­
the number of indicators, the way how it is considered, etc.), this study perature of the HVAC system [27]. According to the “Guidance Notes for
diagnosed the current status of the energy efficiency of the HVAC system the Management of Indoor Air Quality in Offices and Public Places” [39]
in two ways: (i) “basic analysis” - the set-point temperature of the HVAC enacted by Hong Kong, the excellent class of indoor temperature is
system in office spaces by considering the indoor temperature; and (ii) defined within a range of 20–25.5 � C, aiming to create a thermal comfort
“advanced analysis” - the effective utilization of the HVAC system in level of the indoor environment. Meanwhile, the “Energy Saving Charter
office spaces by considering the indoor temperature and CO2 2017" [40] launched by Environment Bureau in Hong Kong rather fo­
concentration. cuses on the energy-saving issue, appealing to maintain an averaged
indoor temperature within a range of 24–26 � C in a summer season (i.e.
2.3.1. Basic analysis from June to September).
The occupants could adjust the set-point temperature of the HVAC By thoroughly considering the aforementioned regulation standards,

4
W. Li et al. Renewable and Sustainable Energy Reviews 127 (2020) 109885

four classes of indoor temperature (i.e. Class 1-Class 4, divided by 20 � C, sensors.


24 � C, and 25.5 � C) were set in this study (refer to SM Table S2). Four In general, a dead band could occur by the relevant properties of
ranges with different characteristics can be used to determine the set- equipment. For example, the relationship between the set-point tem­
point temperature of the HVAC system. perature and the dead band in a cooling mode can be explained as fol­
lows: (i) if the indoor temperature reaches or exceeds the maximum
� Class 1 (higher than 25.5 � C): This class could contribute to the energy temperature (which can be established at the set-point temperature plus
saving in accordance with the regulation standard [40], but the a half of dead band), the HVAC system starts operating in a cooling mode
thermal comfort requirement could not be met with the regulation and (ii) if the indoor temperature reaches or is below the minimum
standard [39]. That is, the energy saving could be achieved by temperature (which can be established at the set-point temperature
turning off the HVAC system in summer; however, the temperature minus a half of dead band), the HVAC system stops operating. In this
could not be satisfied with a thermal comfort level. way, the HVAC system would be operated to maintain the indoor tem­
� Class 2 (24–25.5 � C): This class could meet the regulation standards perature within the dead band (refer to the blue dotted area in Fig. 1);
set by the government in Hong Kong [39,40] in terms of both of and thus, the indoor temperature could be fluctuated around the set-
energy saving and thermal comfort issues. The set-point temperature point temperature. In this background, this study adopted an indirect
of the HVAC system should be adjusted within a range of 24–25.5 � C way to calculate the averaged indoor temperature and regarded it as the
as a preferred standard to simultaneously achieve two goals. set-point temperature of the HVAC system.
� Class 3 (20–24 � C): This class could meet the thermal comfort Meanwhile, the CPA can be adopted to dividing time periods by
requirement set by the government in Hong Kong [39]; however, the considering the set-point temperatures of the HVAC system. The CPA is a
more energy would be consumed to keep the indoor temperature univariate statistical analysis technique to detect the change in the
within a range of 20–24 � C (Class 3) compared to that of 24–25.5 � C statistical properties of a sequence of data [44]. The CPA has been
(Class 2). This class (Class 3) is especially necessary and meaningful applied in many fields, such as fault detection [16], occupancy estima­
for such occupants as children, elderly, and patients, who are very tion [45,46], and the financial exchange rate analysis [47]. The mean
sensitive to the indoor temperature. temperature was regarded as the set-point temperature of the HVAC
� Class 4 (lower than 20 � C): This class could meet neither the energy system in the CPA model. The change point detection process can be
saving nor the thermal comfort issues. If the indoor temperature was illustrated as follows.
set within a range of this class, occupants would feel cold, and po­ The dataset on the indoor temperatures was collected in 15-min in­
tential problems (e.g. condensation [37]) could happen. In other terval by the environmental sensors, and the default operation schedules
words, more energy could be consumed without obtaining occupant’ of the HVAC systems in two office spaces were automatically controlled
satisfaction. Thus, the set-point temperature below 20 � C should be from 7:00 through 23:00. Thus, the 64 datasets of the indoor tempera­
avoided. tures between 7:00 to 23:00 were used as inputs in the CPA model. Let
the time-series data of the indoor temperatures and the data for each
Meanwhile, it is not available to record all the values from the sampling time (15-min interval) denoted by T and Ti, respectively
control panels of the existing HVAC system. Also, the values displayed (where i ¼ 1, 2, …, 64). According to Equation (1), T was separated into
on the control panels could deviate from the actual value controlled by different segments by considering the different change points (θ1, θ2, …,
the HVAC system (i.e. performance gap) due to the equipment failure θj) [45]. The adjacent segments can be featured with various statistical
inside the HVAC system [41–43]. Also, the historical values of the properties (i.e., the mean value of the indoor temperatures for 15-min
set-point temperature of the HVAC system were not recorded. Thus, an interval). Among various CPA algorithms, this study adopted the
alternative way was adopted to estimate the set-point temperature by “changepoint” R package developed by Killick and Eckley [48–51]. The
using the real-time temperature data measured by the environmental function (i.e. “cpt.mean ()") to identify the changes in mean values was

Fig. 1. Application of the CPA model for distinguishing the different set-point temperatures in the large office space (for the case of August 15, 2017).

5
W. Li et al. Renewable and Sustainable Energy Reviews 127 (2020) 109885

chosen, in which the parameters were selected with the Schwartz in­ [54–56]. Even if the mechanical ventilation of the HVAC system is
formation criterion (SIC) as a penalty and the power of the pruned exact being operated so to bring a certain amount of fresh air, the CO2
linear time (PELT) as a method. emissions from occupants could evidently contribute to the increase
8 in the CO2 concentration [57,58]. Thus, as a proxy indicator, the CO2
> T1 ði; w1 Þ þ e1 ðiÞ; 1 � i < θ1
>
< T ði; w Þ þ e ðiÞ; θ < i < θ concentration can be adopted to evaluate whether the office spaces
T¼ 2 2 2 1 2
(1) are being occupied. For example, if the CO2 concentration would
> ……
>
: �
Tj i; wj þ ej ðiÞ; θj 1 < i � 64 increase for more than 30 min, an occupant would be regarded to
have come into an office space (i.e. starting a meeting); otherwise, it
where, Tk (i,wk) is the function to describe the statistical property of the would be regarded to have left an office space (i.e. finishing a
kth segment; i is the sampling time; wk is the parameter vector; and ek(i) meeting).
is the error term of the kth segment, k ¼ 1, 2, …, j.
Fig. 1 shows an example of the CPA model in the large office space on By considering the real-time data of indoor temperature and CO2
Aug. 15, 2017. The indoor temperatures collected in 15-min interval concentration simultaneously, it is possible to judge whether the HVAC
between 7:00 to 23:00 were numbered in order from 1 to 64, and they system would be effectively operated. If the HVAC system would be
were used as the inputs of the function “cpt.mean ()" in the “change­ operated during a meeting, it would be meaningful to keep a comfort­
point” R package. After running the “R" programming language, the able environment. Otherwise, the HVAC system should be turned off so
11th data (i.e. temperature at 9:30) was identified as the change point; as to achieve the potential of the energy efficiency improvement of the
and thus, the dataset between 7:00 to 23:00 were divided into two HVAC system. Fig. 2 shows the time-series indoor temperature and CO2
segments: (i) 7:00–9:15 (the first segment) and (ii) 9:30–23:00 (the concentration in the large office space on Sep. 25, 2017, illustrating the
second segment). The mean value in the period of 7:00–9:15 (the first aforementioned evaluation process.where, EURH is the effective utili­
segment) and 9:30–23:00 (the second segment) were 27.13 � C and zation ratio of the HVAC system (%); ODH is the operation duration of
20.00 � C, respectively. Since the set-point temperature of the HVAC the HVAC system (hour); and DM is the duration of meeting (hour).
system could be adjusted with the increment of 0.5 � C via its control ADM
panel, the calculated mean temperature should be rounded to the first AEURH ¼ (3)
AODH
decimal place with a value of 0 or 5 (e.g., from the calculated mean
temperature of 27.13 � C to the rounded value of 27.0 � C). where, AEURH is the accumulated effective utilization ratio of the HVAC
As explained earlier, there are four classes on the indoor temperature system (%); AODH is the accumulated operation duration of the HVAC
by considering the regulation standards set by the government in Hong system (hour); and ADM is the accumulated duration of meeting (hour).
Kong. Assuming that if the indoor temperature lies in Class 1 (i.e. higher
than 25.5 � C), the HVAC system could be inferred to be turned off. If the
indoor temperature lies in Class 3 (i.e. 20–24 � C), the level of thermal 2.4. Intervention phase
comfort could be satisfied, and the potential of energy savings could be
more increased. In the example shown in Fig. 1, it could be inferred: (i) 2.4.1. Basic control strategy
the HVAC system was not operated between 7:00–9:15 (the first As explained in Section 2.3.1, the function “cpt.mean ()" in the
segment); and (ii) the energy saving could be achieved if the set-point “changepoint” R package was adopted to estimate the set-point tem­
temperature between 9:30–23:00 (the second segment) would be perature of the HVAC system. On one hand, in case of reaching a lower
increased to Class 2 (i.e. 24–25.5 � C). For the whole dataset from 31 Jul. set-point temperature, the indoor temperature would first decrease to
through 30 Sep. in 2017, it can be found in SM Tables S3–S4 (for small the minimum limit of the corresponding set-point temperature (Tmin),
and large office spaces, respectively). and then it would fluctuate within a certain range between “Tmin” and
“Tmin plus dead band”. That is, the basic concept for the real-time esti­
2.3.2. Advanced analysis mation of the lower set-point temperature of the HVAC system is to
Occupants can have an effect on the indoor environment in both identify “Tmin” and a certain temperature that lies between “Tmin” and
ways: (i) passive way (e.g. heat generation, moisture generation, CO2 “Tmin plus dead band”. On the other hand, in case of reaching a higher
emission, and pollutant emission); and (ii) active way (e.g. controlling set-point temperature, the indoor temperature would first increase to
electronic equipment) [52]. To remove these artificial impacts and to the maximum limit of the corresponding set-point temperature (Tmax),
maintain the indoor environmental indicators within a comfortable and then it would fluctuate within a certain range between “Tmax” and
range, it is a reliable and essential way to operate the HVAC system in “Tmax minus dead band”. That is, the basic concept for the real-time
hot and humid regions, such as Hong Kong [53]. However, considering estimation of the higher set-point temperature of the HVAC system is
the gap between the amount of influence from occupants and the ca­ to identify “Tmax” and a certain temperature that lies between “Tmax”
pacity of the HVAC system, it is not easy to identify the exact cause of the and “Tmax minus dead band".
change in the indicators. The main features of the temperature and CO2 In short, a control strategy with the estimated set-point temperature
concentration in a cooling mode of the HVAC system can be explained as of the HVAC system in real time can be developed for: (i) detecting the
follows [54–58], and others (i.e., humidity, VOCs and micro dust) can be potential failures in the HVAC system and (ii) providing feedbacks on
found in SM Table S6 [59–65]. the adjustment of the set-point temperature within the preferred range
(i.e. Class 2, 24–25.5 � C) if necessary.
� Temperature: Compared to the capacity of the HVAC system in a
cooling mode, the increase in the indoor temperature by occupants’ 2.4.2. Advanced control strategy
activities would be a negligible amount. Thus, the variation of the For the office spaces with irregular occupancy schedule, it would be
indoor temperature can be used as a proxy indicator to evaluate so difficult to pre-set the operation schedule of the HVAC system; and
whether the HVAC system would be being operated. For example, if thus, the predictive control would not be applicable for the HVAC sys­
the indoor temperature would decrease for more than 30 min or be tem in office spaces. In addition, the meeting attendees would seldom
kept below 25.5 � C, it is assumed that the HVAC system would be turn off the HVAC system after meeting; and thus, it would be unreliable
operated; otherwise, it would be turned off.. to turn on/off the HVAC system in a manual way. In this regard, the real-
� CO2 concentration: When an office space is being occupied, the CO2 time occupancy-based control strategy for the HVAC system in office
concentration would generally increase due to the metabolic process spaces would be a promising way. According to Equations (2) and (3),
the AEURH and EURH are in proportion to the ADM and DM; but in

6
W. Li et al. Renewable and Sustainable Energy Reviews 127 (2020) 109885

Fig. 2. The time-series indoor temperature and CO2 concentration in the large office space (for the case of Sep. 25, 2017).

� Operation duration of the HVAC system (ODH): This index can be calculated by considering the variation of indoor temperature as a proxy indicator (refer to the
orange-colored solid line in Fig. 2). The indoor temperature started to decrease for more than 30 min since 9:00 a.m., and it has been kept below 22.5 � C until 10:00
a.m. Then, it started to increase for more than 30 min since 10:00 a.m.; but it has been kept below 22.5 � C until 13:00 p.m. After that, it started to decrease for more
than 30 min since 13:00 p.m., and it has been kept below 22.5 � C until 18:00 p.m. Finally, it started to increase since 18:00 p.m. In short, the HVAC system could be
regarded as being operated from 9:00 a.m. through 18:00 p.m. (refer to the black-colored solid arrow line in Fig. 2); and thus, the ODH can be calculated as 9.0 h.
� Duration of meeting (DM): As a proxy indicator, this index can be calculated by investigating the variations of CO2 concentration. As shown in Fig. 2 (blue-colored
shared area), it could be assumed to have meetings in three time periods: (i) from 9:00 a.m. to 12:00 p.m. (3 h); (ii) from 14:00 p.m. to 16:00 p.m. (2 h); and (iii)
from 17:30 p.m. to 18:00 p.m. (0.5 h). Thus, the DM can be calculated as 5.5 h.
� Effective utilization ratio of the HVAC system (EURH): This index can be determined by considering the ODH and DM (refer to Equation (2)). The less the EURH would
be, the more energy would be wasted (i.e. larger potential of the energy efficiency improvement of the HVAC system). If the EURH would be at 100%, the HVAC
system would be regarded to be operated only during the meeting periods, indicating that it would be most effectively operated without any waste. As a result, the
EURH can be calculated at 61.11% (¼ 5.5 h/9.0 h) by considering two values (i.e. ODH and DM).
� Accumulated effective utilization ratio of the HVAC system (AEURH): This index can be calculated from a macro perspective (refer to Equation (3)). By accumulating
the values of the ODH and DM, the accumulated operation duration of the HVAC system (AODH) and the accumulated duration of meeting (ADM) can be obtained,
respectively. With these values, the AEURH can be calculated (e.g., 40.29% (¼ 213.25 h/529.25 h)), indicating a large potential of the energy efficiency
improvement of the HVAC system in the large office space (e.g., 59.71%).
DM
EURH ¼ (2)
ODH

inversely proportion to the AODH and ODH. Since the ADM and DM 3.1. Basic analysis and control
would be difficult to be predicted and controlled, the AODH and ODH
were determined as target parameters. That is, the HVAC system should 3.1.1. Basic analysis
be properly operated only during the meeting periods, and it should be Fig. 3 shows the distribution of the indoor temperature in two office
turned off when there is no occupant. In this way, the AEURH and EURH spaces in accordance with four classes set by the regulation standards in
could be optimized to the value of 100% (i.e. completely effective uti­ Hong Kong [39,40]. As explained in SM Table S2, four classes have
lization of the HVAC system). different characteristics in terms of energy saving and thermal comfort,
In short, the occupancy-based control strategy would be a reactive this study analyzed the distribution of the indoor temperature by class.
control one, which could respond to the absence of occupants in real
time by investigating the CO2 concentration. Further, this approach � Class 1 (higher than 25.5 � C): The proportions were determined at
could consider the occupants’ preference and behaviors, because it 30.46% and 33.50% for the small and large office spaces. During the
would be promised for the HVAC system to be operated during the meetings, it is required to take the thermal comfort into account; and
meeting periods. thus, an automatic control strategy should be adopted to turn on the
HVAC system in a timely manner.
3. Results and discussion � Class 2 (24–25.5 � C): The smallest proportions were determined at
17.08% and 6.87% for the small and large office spaces, indicating
This study used the real-time indoor environmental indicators to that it would be possible to simultaneously achieve both of energy
analyze the potential of the energy efficiency improvement of the HVAC saving and thermal comfort.
system in office spaces. The detailed analyses were conducted, and the � Class 3 (20–24 � C): The largest proportions were determined at
relevant control strategies were developed in two ways. 47.10% and 41.34% for the small and large office spaces, in which
there would be huge potential for energy saving if the indoor tem­
perature could be adjusted within the range of Class 2 (i.e. 24-25.5

7
W. Li et al. Renewable and Sustainable Energy Reviews 127 (2020) 109885

Fig. 3. Distribution of the indoor temperature by four classes in two office spaces.

C), especially to the upper limit of Class 2 (i.e. 25.5 � C). In addition,

temperature of the HVAC system could be regarded to increase to a
thermal comfort would still be satisfied in this class. higher value, and the step (ii-2) will be followed.
� Class 4 (lower than 20 � C): The proportions were determined at � (ii-1/ii-2) Calculation of the mean value of the indoor temperature:
5.36% and 18.29% for the small and large office spaces, indicating Depending on the judgement in step (i), the way how the set-point
that the set-point temperature would be set at a very low level temperature is estimated could be determined. The basic concept
(especially for the large office space). This class should be entirely of step (ii-1) is to identify “Tmin” and to calculate the mean value of
avoided because of uncomfortable cold feeling, energy waste, and temperatures that lie between “Tmin” and “Tmin plus dead band”.
dew condensation. First, T(1) is initially set as “Tmin”. As the sampling time increases,
“Tmin” will be replaced repeatedly by the temperatures lower than
Based on these analyses, it can be said that the indoor temperature “Tmin” (i.e. “Tmin candidates”). As shown in Fig. 1, the dead band of
needs to be properly maintained within the prescribed range so that both the HVAC system was determined at 2 � C; and thus, the temperature
of energy saving and thermal comfort could be simultaneously managed. between “Tmin” and “Tmin þ 2 � C00 will be considered to calculate the
To do so, it is necessary to firstly understand the current status of the set- mean value (“Tm”). In the process, “i" is the cycle counter to count the
point temperature of the HVAC system. As shown in SM Tables S3–S4, sampling times; “j" is the cycle counter to count the number of sta­
the set-point temperature of the HVAC system was estimated within a bilized data between “Tmin” and “Tmin þ 2 � C"; and “k" is the cycle
range of 19.5–27.5 � C for the small office space and 17.5–29.0 � C for the counter to count the number of “Tmin candidate”. “Tmin” could be
large office space. Also, Fig. 4 shows the frequency of the estimated set- determined in a cyclic process, in which the equality between i and k
point temperature of the HVAC system in two office spaces, in which the is used to evaluate whether to proceed to the next process. In other
median value of the set-point temperature was calculated to clearly words, if the data collected in the current sampling time would be
evaluate the skewness of the distribution. They were determined at 22.0 lower than “Tmin” after the stabilization period of the indoor tem­

C and 21.0 � C (which belonged to Class 3; 20–24 � C) for the small and perature (i.e. between “Tmin” and “Tmin þ 2 � C00 ), it would replace
large office spaces, indicating that occupants would tend to designate “Tmin” but not be treated as “Tmin candidate”. This is because the
the lower set-point temperatures for a cooler indoor environment, and indoor temperature should decrease to “Tmin” directly after estab­
they would neglect the energy-saving aspect in two rooms. lishing the set-point temperature; and then, it would fluctuate within
Therefore, it is required to achieve the energy savings by deploying a the dead band. As for the step (ii-2), the basic concept is to identify
real-time automatic control process, which could estimate the set-point “Tmax” and to calculate the mean value of temperatures that lie be­
temperature of the HVAC system and adjust it to Class 2 (24–25.5 � C, tween “Tmax” and “Tmax minus dead band”, which is very similar to
especially to the upper limit of 25.5 � C) in a timely manner. step (ii-1).
� (iii) Estimation of the set-point temperature: If the number of stabilized
3.1.2. Basic control strategy data (“j") is no less than the value of 2, it indicates that the indoor
Fig. 5 shows a logical flowchart for estimating the set-point tem­ temperature has been within the dead band (i.e., stabilized change)
perature of the HVAC system in real time, which can be applied to (i) for at least 30 min. In this case, the mean value (“Tm”), rounded to
anomaly detection of the HVAC system and (ii) suggestion of the the first decimal place with the value of 0 or 5, would be regarded as
adjustment of the set-point temperature to the preferred range (e.g. the estimated set-point temperature of the HVAC system (“TEs").
Class 2 of 24–25.5 � C). A basic control strategy can be deployed in the � (iv) Anomaly detection of the HVAC system: The estimated set-point
following five steps. temperature (“TEs”) could be compared with the designated set-
point temperature displayed on control panel (“Ts”). If the differ­
� (i) Judgement of whether the set-point temperature decreases: If the data ence between “TEs” and “Ts” would be within a certain level (e.g. 0.5
declines from the initial sampling time (T(0)) through the first sam­ �
C), the HVAC system could be considered to be operated in a normal
pling time (T(1)), it results that the set-point temperature of the condition. Otherwise, it could be assumed to have any of equipment
HVAC system could be regarded to decrease to a lower value for the failures (e.g. bias of the measured data in a sensor, or signal errors)
last 15 min, and the step (ii-1) will be followed to estimate the set- inside the HVAC system. In this case, it should be recommended to
point temperature in real time. Otherwise, the set-point check and calibrate the working condition of the HVAC system.

8
W. Li et al. Renewable and Sustainable Energy Reviews 127 (2020) 109885

Fig. 4. Frequency of the estimated set-point temperature in two office spaces.

� (v) Suggestion of the adjustment of the set-point temperature: If the 3.2. Advanced analysis and control
estimated set-point temperature (“TEs”) is maintained below a
certain range, it could be recommended to adjust the set-point 3.2.1. Advanced analysis
temperature within the range of 24–25.5 � C (e.g., Class 2). Table 3 and SM Table S5 show the “advanced analysis” for the
effective utilization of the HVAC system in two office spaces. Consid­
ering that the HVAC system is automatically operated by the BEMS, the

9
W. Li et al. Renewable and Sustainable Energy Reviews 127 (2020) 109885

Fig. 5. A logical flowchart for estimating the set-point temperature of the HVAC system.

shutdown periods (i.e. from 23:00 p.m. to 7:00 a.m. in working days as meeting until the manual turn-off at 14:15 p.m. in 2 August (refer to the
well as the whole day in holidays) were not included in calculating the shaded area of Table 3). As a result, the ODH was determined at 21.75 h,
ODH (refer to the third column in Table 3 and SM Table S5). For which was summed up by 14.5 h (from 8:30 a.m. to 23:00 p.m.) and
example, the HVAC system in the large office space would start to 7.25 h (from 7:00 a.m. to 14:15 p.m.). The results can be described in
operate at 8:30 a.m. on 1 August, and it would be kept operating despite detail from the different perspectives as follows:
no meeting until the automatic shutdown by the BEMS at 23:00 p.m. in First, on the macro level, even if the AEURH would be determined at
the same day. Then, the HVAC system would be turned on automatically a very small value, it could be expected to have a large potential of the
by the BEMS at 7:00 a.m. in 2 August, and it would be kept despite no energy efficiency improvement of the HVAC system in two office spaces.

10
W. Li et al. Renewable and Sustainable Energy Reviews 127 (2020) 109885

Table 3
“Advanced analysis” for the effective utilization of the HVAC system in the large office space.
Starting time of the HVAC system (manually controlled) Downtime of the HVAC system (manually controlled) ODHa (h) Meeting Time DMb (h) EURHc (%)

Jul.31st,8:30 Jul.31st,19:45 11.25 Jul.31st,8:30–9:45 5.75 51.11


Jul.31st,11:30–13:45
Jul.31st,14:15–16:30
Aug.1st,8:30 Aug.2nd,14:15 21.75 Aug.1st,8:30–10:30 8.25 37.93
Aug.1st,11:45–12:45
Aug.1st,14:15–16:45
Aug.2nd,7:15–10:30
Aug.3rd,9:30 Aug.9th,14:15 72.75 Aug.3rd,9:30–11:15 22.5 30.93
Aug.3rd,13:30–15:15
Aug.4th,7:45–10:45
Aug.7th,7:30–11:45
Aug.8th,7:45–12:00
Aug.8th,14:30–16:15
Aug.8th,16:30–17:15
Aug.9th,7:45–11:45
Aug.9th,12:15–13:15
Aug.10th,8:45 Aug.14th,19:45 43 Aug.10th,8:45–12:30 20.75 48.26
Aug.11th,7:30–10:30
Aug.11th,14:00–15:15
Aug.11th,16:30–18:00
Aug.11th,20:00–21:00
Aug.14th,7:15–13:15
Aug.14th,14:00–15:30
Aug.14th,16:00–17:45
Aug.15th,9:15 Aug.29th,19:15 170 Aug.15th,9:15–10:45 66.5 39.12
Aug.15th,13:45–15:00
Aug.15th,15:45–16:30
Aug.15th,18:15–20:30
Aug.16th,8:00–11:00
Aug.16th,14:00–15:15
Aug.17th,8:00–12:15
Aug.17th,14:15–15:00
Aug.17th,15:15–16:00
Aug.17th,17:45–18:45
Aug.18th,8:00–12:00
Aug.18th,13:30–17:00
Aug.21st,8:00–12:45
Aug.21st,14:00–16:45
Aug.21st,17:00–18:00
Aug.22nd,8:00–10:45
Aug.22nd,11:00–12:45
Aug.22nd,14:15–16:30
Aug.22nd,16:45–18:30
Aug.24th,8:45–11:15
Aug.24th,13:45–17:15
Aug.25th,9:00–13:00
Aug.25th,14:00–15:30
Aug.25th,16:00–17:15
Aug.25th,18:00–19:00
Aug.28th,8:00–11:30
Aug.28th,14:15–17:30
Aug.29th,7:30–9:30
Aug.29th,14:30–17:15

Starting time of the HVAC system (manually controlled) Downtime of the HVAC system (manually controlled) ODHa (h) Meeting Time DMb (h) EURHc (%)

Aug.30th,12:00 Aug.30th,19:15 7.25 Aug.30th,12:00–13:45 3.75 51.72


Aug.30th,14:15–15:00
Aug.30th,16:30–17:45
Sep.1st,9:30 Sep.1st,11:45 2.25 Sep.1st,9:30–10:45 1.25 55.56
Sep.4th,9:00 Sep.7th,23:00d 62 Sep.4th,9:00–10:15 22.75 36.69
Sep.4th,12:30–13:45
Sep.4th,15:00–16:00
Sep.4th,16:45–17:45
Sep.5th,7:45–11:15
Sep.5th,11:30–12:15
Sep.5th,14:00–15:15
Sep.5th,16:45–17:30
Sep.6th,7:45–10:15
Sep.6th,11:00–12:00
Sep.6th,14:15–15:45
Sep.7th,8:00–10:45
Sep.7th,14:15–16:00
Sep.7th,17:00–19:30
Sep.8th,9:15 Sep.8th,21:45 12.5 Sep.8th,9:15–12:00 6 48.00
(continued on next page)

11
W. Li et al. Renewable and Sustainable Energy Reviews 127 (2020) 109885

Table 3 (continued )
Starting time of the HVAC system (manually controlled) Downtime of the HVAC system (manually controlled) ODHa (h) Meeting Time DMb (h) EURHc (%)

Sep.8th,14:15–17:30
Sep.11th,12:45 Sep.11th,15:45 3 Sep.11th,12:45–14:00 1.25 41.67
Sep.12th,15:15 Sep.12th,21:15 6 Sep.12th,15:15–18:30 3.25 54.17
Sep.13th,9:45 Sep.13th,10:45 1 Sep.13th,9:45–10:45 1 100.00
Sep.18th,7:00 Sep.19th,19:00 28 Sep.18th,8:15–10:00 7.5 26.79
Sep.18th,11:30–12:15
Sep.18th,15:30–16:45
Sep.19th,8:15–11:00
Sep.19th,11:15–12:15
Sep.20th,11:00 Sep.21st,14:15 19.25 Sep.20th,11:00–12:00 8.5 44.16
Sep.20th,14:45–16:15
Sep.20th,16:30–17:30
Sep.20th,18:00–20:00
Sep.21st,8:45–11:45
Sep.21st,15:30 Sep.21st,21:00 5.5 Sep.21st,15:30–18:00 2.5 45.45
Sep.22nd,9:30 Sep.22nd,13:00 3.5 Sep.22nd,9:30–10:45 1.25 35.71
Sep.22nd,15:30 Sep.22nd,20:30 5 Sep.22nd,15:30–18:15 2.75 55.00
Sep.25th,9:00 Sep.25th,20:15 9 Sep.25th,9:00–12:00 5.5 61.11
Sep.25th,14:00–16:00
Sep.25th,17:30–18:00
Sep.26th,9:15 Sep.26th,20:00 10.75 Sep.26th,9:15–13:00 6.75 62.79
Sep.26th,14:00–16:15
Sep.26th,17:00–17:45
Sep.27th,9:15 Sep.27th,19:45 10.5 Sep.27th,9:15–12:30 5 47.62
Sep.27th,14:00–15:45
Sep.28th,10:00 Sep.28th,12:15 2.25 Sep.28th,10:00–11:30 1.5 66.67
Sep.28th,16:15 Sep.29th,23:00d 22.75 Sep.28th,16:30–17:45 9 39.56
Sep.29th,8:00–11:00
Sep.29th,11:15–12:30
Sep.29th,12:45–13:45
Sep.29th,14:30–15:45
Sep.29th,16:30–17:45
AODHe (h) 529.25 ADMf (h) 213.25
AEURHg (%) 40.29
a b c
Notes: ODH stands for the operation duration of the HVAC system; DM stands for the duration of meeting; EURH stands for the effective utilization ratio of the
HVAC system; d Part of data were missed. It was not easy to infer when the HVAC system would be turned off manually; thus, 23:00 p.m. (which is automatic shutdown
time) was listed here; e AODH stands for the accumulated operation duration of the HVAC system; f ADM stands for the accumulated duration of meeting; and g AEURH
stands for the accumulated effective utilization ratio of the HVAC system.

In other words, the ADM and AODH were determined at 99.75 and small office space and 26.79–66.67% for the large office space. It
220.25 h for the small office space and at 214.25 and 531.5 h for the means that the meeting attendees would forget turning off the HVAC
large office space. The large difference between the ADM and AODH system and make it continue to work for a long period after the
indicates that the HVAC system would not be effectively operated in the meeting. As a result, the HVAC system has not been properly oper­
office spaces. As a result, the AEURH was determined at 45.29% for the ated, indicating that 41.38–84.00% and 33.33–73.21% of energy
small office space and at 40.29% for the large office space. For the small consumption by the HVAC system would be wasted in the small and
office space, the amount of energy consumed by the HVAC system would large office spaces despite no person in there.
be suitably used at a level of 45.29% to provide meeting attendees with a
comfortable indoor temperature, while 54.71% of the energy con­ Therefore, considering the accumulated value (i.e. AEURH) or indi­
sumption in a cooling mode would be wasted in an unoccupied room. vidual value (i.e. EURH) in two office spaces, it is possible to achieve a
For the large office space, only 40.29% of the cooling energy supplied by large potential of the energy efficiency improvement of the HVAC sys­
the HVAC system would be meaningful, and 59.71% could be potentially tem. To do so, the automatic control strategy for the real-time occu­
saved. pancy-based shutdown control of the HVAC system can be implemented.
Second, on the micro level, it was determined that the two office
spaces would be irregularly occupied, and the HVAC system would not 3.2.2. Advanced control strategy
be timely controlled before and after the meetings, leading to energy Fig. 6 shows a logical flowchart for the automatic on/off control of
waste. The detailed results were explained as follows. the HVAC system in real time, which can be applied to improve the
effective utilization of the HVAC system. The main concept is to timely
� The meeting schedule: Compared to a regularly occupied space in a turn off the HVAC system after the meeting by considering the CO2
normal nine-to-five working mode, the two office spaces were concentration. An advanced control strategy can be deployed in the
determined to be irregularly occupied. Thus, it would be very diffi­ following five steps.
cult to define the meeting schedule and to set up the designated
operation schedule of the HVAC system in advance. � (i) Data collection: As shown in SM Table S1 (specifications of the
� The operation of the HVAC system in office spaces: Since the HVAC indoor environmental sensor), the dataset collected in 1-min interval
system in an office space could be controlled in a manual mode, it on the CO2 concentration (“CO(i)2 ") and indoor temperature (“T ")
(i)

would be often excessively operated so that a large amount of cooling can be used to develop the advanced control strategy, in which “i" is
energy could be wasted. As shown in Table 3 and SM Table S5, in the cycle counter to count the sampling times.
most of cases, the DM was evidently smaller than the ODH, of which
ratio (i.e. EURH) was ranged between 16.00 and 58.62% for the

12
W. Li et al. Renewable and Sustainable Energy Reviews 127 (2020) 109885

Fig. 6. A logical flowchart for the automatic on/off control of the HVAC system in real time.

� (ii) Judgment of whether the CO2 concentration decreases: If the CO2 decrease or be kept at a low level for 15 min. That is, it can be judged
concentration would decrease for 1 min, step (iii) will be conducted. that the HVAC system would keep operating after the meetings. In
Otherwise, the process will be restarted for the first step. this case, step (v) will be conducted. Otherwise, the process will be
� (iii) Judgment of whether the indoor temperature decreases or keeps at a restarted for the first step.
low level: If the indoor temperature would be below 25.5 � C (i.e., the � (v) Automatic on/off control of the HVAC system: This step would be
upper limit of Class 2), the indoor temperature would be kept at a conducted only when the conditions in step (iv) could be satisfied,
low level. If the aforementioned condition would happen, step (iv) indicating that occupants would forget to turn off the HVAC system
will be conducted. Otherwise, the process will be restarted for the and the HVAC system should be turned off automatically.
first step.
� (iv) Judgment of whether the HVAC system keeps operating after the
3.3. Discussion
meeting: This step could be conducted only when the conditions in
steps (ii) and (iii) could be satisfied. It means that the CO2 concen­
The real-time automatic control strategies proposed in this study can
tration would decrease for 15 min and the indoor temperature would
be utilized in connection with renewable and sustainable energy systems

13
W. Li et al. Renewable and Sustainable Energy Reviews 127 (2020) 109885

in buildings as follows. 40.29% for the large office space, indicating that 54.71% of the cool
energy consumption for the small office space would be wasted in an
� Demand response: The basic control strategy can be linked to the unoccupied mode; and 59.71% for the large office space. That is, it
demand response program in buildings. That is, it is possible to could be judged that the HVAC system would be kept operating since
control the set-point temperature of the HVAC system in more ac­ occupants would forget to turn off the HVAC system after the
curate manner so as to avoid the peak load at a specific point in time. meetings. To solve this problem, this study proposed a logical control
Through this approach, it is expected to create an opportunity to strategy for the automatic on/off control of the HVAC system in real
receive incentives from the government according to demand time, which can be used to timely turn off the HVAC system after the
response program. meetings by considering the variation of the temperature and CO2
� Distributed generation system: The advanced control strategy can be concentration in real time
linked to the distributed generation system in buildings. That is,
irregularly occupied space (e.g., meeting room) can be connected to In conclusion, this study can be a promising attempt for achieving
distributed generation systems (e.g. building-integrated photovol­ the energy efficiency improvement of the HVAC system in irregularly
taic, fuel cell, etc.), which provides energy to a space once it is occupied offices (i.e. office spaces) by utilizing real-time big data.
occupied. Also, the potential energy savings can be utilized to Particularly, the occupancy in a room could be detected by the variation
operate a fuel cell system. of the CO2 concentration in a non-intrusive manner. Furthermore, it is
expected to have a large amount of energy saving potential by effec­
Meanwhile, the following issues can be treated to further improve tively operating the HVAC system in irregularly occupied rooms, where
the proposed real-time automatic control strategies in the follow-up the traditional centralized control system (i.e. BEMS) with a fixed
studies. schedule and manual control mode would not be appropriate.

� Basic control strategy: It is needed to consider the subjective percep­ Declaration of competing interest
tion of occupants and the adaptability to other districts with more
comprehensive classes. Also, it is required to determine the threshold The authors declare no competing financial interest.
level of the number of stabilized data (“j") in step (iii) of Fig. 5 as well
as the threshold level of difference between the estimated set-point
temperature (“TEs”) and the designated one displayed on control CRediT authorship contribution statement
panel (“Ts”) in step (iv) of Fig. 5.
� Advanced control strategy: To further improve the accuracy of the Wenzhuo Li: Investigation, Data curation, Writing - original draft.
automatic on/off control of the HVAC system, it is suggested to Choongwan Koo: Supervision, Conceptualization, Funding acquisition,
determine the threshold level of the low temperature level in step Writing - review & editing. Taehoon Hong: Conceptualization, Project
(iii) of Fig. 6 and the cycle counter value (i) in step (iv) of Fig. 6 by administration, Funding acquisition. Jeongyoon Oh: Conceptualiza­
considering the project characteristics and the relevant properties tion, Resources, Validation. Seung Hyun Cha: Conceptualization,
and by calibrating the control strategy with the ground truth. Formal analysis. Shengwei Wang: Conceptualization, Methodology.

4. Conclusions Acknowledgements

This study aimed to develop a novel operation approach for the en­ This work was supported by the National Research Foundation of
ergy efficiency improvement of the HVAC system in office spaces, in Korea (NRF) grant funded by the Korean government (MSIP; Ministry of
which the real-time big data analytics were adopted. The living-lab Science, ICT & Future Planning) (NRF-2020R1C1C1004147).
approach was used to conduct the field measurement for two office
spaces (i.e. the large and small office spaces) in “A" office located in Appendix A. Supplementary data
Hong Kong from 31 Jul. through 30 Sep. in 2017. The real-time indoor
environmental indicators were used to analyze the current operation Supplementary data to this article can be found online at https://doi.
status of the HVAC system and to propose the automatic control stra­ org/10.1016/j.rser.2020.109885.
tegies in the following two ways.
References
(i) Basic control strategy: The set-point temperature of the HVAC
system was estimated by the CPA model that could identify the [1] National Development and Reform Commission. Comprehensive work plan for
mean value of the indoor temperature in real time. Then, the conserving energy and reducing emissions for the thirteenth five-year plan.
median value of the estimated set-point temperature was calcu­ National Development and Reform Commission; 2016. 2019, http://www.ndrc.gov
.cn/zcfb/zcfbqt/201701/W020170105634585914832.pdf.
lated to clearly evaluate the skewness of the distribution, result­ [2] Ouf M, Issa M, Merkel P. Analysis of real-time electricity consumption in Canadian
ing at 22.0 � C and 21.0 � C for the small and large office spaces, school buildings. Energy Build 2016;128:530–9.
respectively. It could be stated that occupants would tend to set [3] The electrical & mechanical services department of the government of the Hong
Kong special administrative region. Hong Kong Energy End-use Data 2017; 2017.
up the lower set-point temperatures for a cooler indoor envi­ [4] Gul MS, Patidar S. Understanding the energy consumption and occupancy of a
ronment, indicating that a large amount of cooling energy could multi-purpose academic building. Energy Build 2015;87:155–65.
be wasted. To solve this problem, this study proposed a logical [5] Katipamula S, Underhill RM, Goddard JK, Taasevigen DJ, Piette M, Granderson J,
Brown RE, Lanzisera SM, Kuruganti T. Small-and medium-sized commercial
control strategy for estimating the set-point temperature of the building monitoring and controls needs: a scoping study, Pacific Northwest
HVAC system in real time, which can be used to conduct the National Lab.(PNNL). 2012. Richland, WA (United States).
anomaly detection of the HVAC system and to automatically [6] Energy Information Administration. How many smart meters are installed in the
United States, and who has them?. 2016. 2018, https://www.eia.gov/tools/faqs/fa
adjust the set-point temperature within a preferred range.
q.php?id¼108&t¼1.
(ii) Advanced control strategy: The effective utilization level of the [7] Miller C, Nagy Z, Schlueter A. A review of unsupervised statistical learning and
HVAC system was evaluated by comparing the meeting schedule and visual analytics techniques applied to performance analysis of non-residential
the operation of the HVAC system in office spaces. On a macro level, buildings. Renew Sustain Energy Rev 2018;81:1365–77.
[8] Chou J-S, Ngo N-T. Time series analytics using sliding window metaheuristic
the accumulated effective utilization ratio of the HVAC system optimization-based machine learning system for identifying building energy
(AEURH) was determined at 45.29% for the small office space and at consumption patterns. Appl Energy 2016;177:751–70.

14
W. Li et al. Renewable and Sustainable Energy Reviews 127 (2020) 109885

[9] Luo X, Hong T, Chen Y, Piette MA. Electric load shape benchmarking for small- and dynamic energy performance: focused on the operation and maintenance phase.
medium-sized commercial buildings. Appl Energy 2017;204:715–25. Appl Energy 2015;155:671–707.
[10] An J, Yan D, Hong T. Clustering and statistical analyses of air-conditioning [37] Li W, Koo C, Cha SH, Hong T, Oh J. A novel real-time method for HVAC system
intensity and use patterns in residential buildings. Energy Build 2018;174:214–27. operation to improve indoor environmental quality in meeting rooms. Build
[11] Ma Z, Yan R, Nord N. A variation focused cluster analysis strategy to identify Environ 2018;144:365–85.
typical daily heating load profiles of higher education buildings. Energy 2017;134: [38] De Mauro A, Greco M, Grimaldi M. A formal definition of Big Data based on its
90–102. essential features. Libr Rev 2016;65(3):122–35.
[12] Ozawa A, Furusato R, Yoshida Y. Determining the relationship between a [39] Indoor air quality management group of the government of the Hong Kong special
household’s lifestyle and its electricity consumption in Japan by analyzing administrative region, guidance Notes for the management of indoor air quality in
measured electric load profiles. Energy Build 2016;119:200–10. offices and public places, 2003.
[13] Yang J, Ning C, Deb C, Zhang F, Cheong D, Lee SE, Sekhar C, Tham KW. k-Shape [40] Environment Bureau in Hong Kong. Energy saving charter 2017. http://www.
clustering algorithm for building energy usage patterns analysis and forecasting energysaving.gov.hk/esc2017/en/home/index.html; 2017.
model accuracy improvement. Energy Build 2017;146:27–37. [41] Du Z, Fan B, Chi J, Jin X. Sensor fault detection and its efficiency analysis in air
[14] Chou J-S, Telaga AS. Real-time detection of anomalous power consumption. Renew handling unit using the combined neural networks. Energy Build 2014;72:157–66.
Sustain Energy Rev 2014;33:400–11. [42] Schein J, Bushby ST, Castro NS, House JM. A rule-based fault detection method for
[15] Chou J-S, Telaga AS, Chong WK, Gibson GE. Early-warning application for real- air handling units. Energy Build 2006;38(12):1485–92.
time detection of energy consumption anomalies in buildings. J Clean Prod 2017; [43] Borgstein EH, Lamberts R, Hensen JLM. Mapping failures in energy and
149:711–22. environmental performance of buildings. Energy Build 2018;158:476–85.
[16] Horrigan M, Turner WJN, O’Donnell J. A statistically-based fault detection [44] Guralnik V, Srivastava J. Event detection from time series data. In: Proceedings of
approach for environmental and energy management in buildings. Energy Build the fifth ACM SIGKDD international conference on Knowledge discovery and data
2018;158:1499–509. mining. ACM; 1999. p. 33–42.
[17] Li Y, Liu M, Lau J, Zhang B. Experimental study on electrical signatures of common [45] Szczurek A, Maciejewska M. Detection of occupancy events from indoor air
faults for packaged DX rooftop units. Energy Build 2014;77:401–15. monitoring data, Mathematics and Computers in Sciences and in Industry (MCSI).
[18] Fan C, Xiao F, Wang S. Development of prediction models for next-day building In: 2016 third international conference on. IEEE; 2016. p. 229–34.
energy consumption and peak power demand using data mining techniques. Appl [46] Pereira PF, Ramos NMM. Detection of occupant actions in buildings through
Energy 2014;127:1–10. change point analysis of in-situ measurements. Energy Build 2018;173:365–77.
[19] Afram A, Janabi-Sharifi F, Fung AS, Raahemifar K. Artificial neural network (ANN) [47] Zeileis A, Shah A, Patnaik I. Testing, monitoring, and dating structural changes in
based model predictive control (MPC) and optimization of the HVAC systems: a exchange rate regimes. Comput Stat Data Anal 2010;54(6):1696–706.
state of the art review and case study of a residential HVAC system. Energy Build [48] Killick R, Eckley I. changepoint: an R package for changepoint analysis. J Stat
2017;141:96–113. Software 2014;58(3):1–19.
[20] Ghofrani A, Jafari MA. Distributed air conditioning control in commercial [49] Killick R, Eckley I. Changepoint: analysis of changepoint models. Lancaster, UK:
buildings based on a physical-statistical approach. Energy Build 2017;148:106–18. Lancaster University; 2010. http://CRAN.R-project.org/pacakge¼changepoint.
[21] Kuroha R, Fujimoto Y, Hirohashi W, Amano Y, Tanabe S-i, Hayashi Y. Operation [50] Killick R, Fearnhead P, Eckley IA. Optimal detection of changepoints with a linear
planning method for home air-conditioners considering characteristics of computational cost. J Am Stat Assoc 2012;107(500):1590–8.
installation environment. Energy Build 2018;177:351–62. [51] Killick R, Haynes K, Eckley I, Fearnhead P, Lee J. Package ‘changepoint’, R package
[22] Ghahramani A, Zhang K, Dutta K, Yang Z, Becerik-Gerber B. Energy savings from version 0.4.-2011. 2016. http://cran.rproject.org/web/packages/changepoint/in
temperature setpoints and deadband: quantifying the influence of building and dex.html.
system properties on savings. Appl Energy 2016;165:930–42. [52] Calì D, Matthes P, Huchtemann K, Streblow R, Müller D. CO 2 based occupancy
[23] Hoyt T, Arens E, Zhang H. Extending air temperature setpoints: simulated energy detection algorithm: experimental analysis and validation for office and residential
savings and design considerations for new and retrofit buildings. Build Environ buildings. Build Environ 2015;86:39–49.
2015;88:89–96. [53] Ai ZT, Mak CM, Cui DJ, Xue P. Ventilation of air-conditioned residential buildings:
[24] Labeodan T, De Bakker C, Rosemann A, Zeiler W. On the application of wireless a case study in Hong Kong. Energy Build 2016;127:116–27.
sensors and actuators network in existing buildings for occupancy detection and [54] Zuraimi MS, Pantazaras A, Chaturvedi KA, Yang JJ, Tham KW, Lee SE. Predicting
occupancy-driven lighting control. Energy Build 2016;127:75–83. occupancy counts using physical and statistical Co 2 -based modeling
[25] Wang W, Wang J, Chen J, Huang G, Guo X. Multi-zone outdoor air coordination methodologies. Build Environ 2017;123:517–28.
through Wi-Fi probe-based occupancy sensing. Energy Build 2018;159:495–507. [55] Yang J, Santamouris M, Lee SE. Review of occupancy sensing systems and
[26] Wang L, Greenberg S, Fiegel J, Rubalcava A, Earni S, Pang X, Yin R, Woodworth S, occupancy modeling methodologies for the application in institutional buildings.
Hernandez-Maldonado J. Monitoring-based HVAC commissioning of an existing Energy Build 2016;121:344–9.
office building for energy efficiency. Appl Energy 2013;102:1382–90. [56] Szczurek A, Maciejewska M, Wyłoma� nska A, Zimroz R, Zak
_ G, Dolega A. Detection
[27] Yamtraipat N, Khedari J, Hirunlabh J, Kunchornrat J. Assessment of Thailand of occupancy profile based on carbon dioxide concentration pattern matching.
indoor set-point impact on energy consumption and environment. Energy Pol Measurement 2016;93:265–71.
2006;34(7):765–70. [57] Chen Z, Jiang C, Xie L. Building occupancy estimation and detection: a review.
[28] Ghahramani A, Jazizadeh F, Becerik-Gerber B. A knowledge based approach for Energy Build 2018;169:260–70.
selecting energy-aware and comfort-driven HVAC temperature set points. Energy [58] Candanedo LM, Feldheim V. Accurate occupancy detection of an office room from
Build 2014;85:536–48. light, temperature, humidity and CO 2 measurements using statistical learning
[29] Kim W, Katipamula S, Lutes R. Improving HVAC operational efficiency in small- models. Energy Build 2016;112:28–39.
and medium-size commercial buildings. Build Environ 2017;120:64–76. [59] Luoma M, Batterman SA. Characterization of particulate emissions from occupant
[30] Moon JW, Han S-H. Thermostat strategies impact on energy consumption in activities in offices. Indoor Air 2001;11(1):35–48.
residential buildings. Energy Build 2011;43(2–3):338–46. [60] R€osch C, Kohajda T, R€ oder S, Bergen Mv, Schlink U. Relationship between sources
[31] Manjarres D, Mera A, Perea E, Lejarazu A, Gil-Lopez S. An energy-efficient and patterns of VOCs in indoor air. Atmospheric Pollution Research 2014;5(1):
predictive control for HVAC systems applied to tertiary buildings based on 129–37.
regression techniques. Energy Build 2017;152:409–17. [61] Szczurek A, Dolega A, Maciejewska M. Profile of occupant activity impact on
[32] Moon JW, Jung SK. Development of a thermal control algorithm using artificial indoor air — method of its determination. Energy Build 2018;158:1564–75.
neural network models for improved thermal comfort and energy efficiency in [62] Guo M, Pei X, Mo F, Liu J, Shen X. Formaldehyde concentration and its influencing
accommodation buildings. Appl Therm Eng 2016;103:1135–44. factors in residential homes after decoration at Hangzhou, China. J Environ Sci
[33] Yang Z, Becerik-Gerber B. The coupled effects of personalized occupancy profile 2013;25(5):908–15.
based HVAC schedules and room reassignment on building energy use. Energy [63] Hori H, Ishimatsu S, Fueta Y, Ishidao T. Evaluation of a real-time method for
Build 2014;78:113–22. monitoring volatile organic compounds in indoor air in a Japanese university.
[34] Nagarathinam S, Doddi H, Vasan A, Sarangan V, Venkata Ramakrishna P, Environ Health Prev Med 2013;18(4):285–92.
Sivasubramaniam A. Energy efficient thermal comfort in open-plan office [64] Caron A, Hanoune B, Redon N, Coddeville P. Gas sensor networks: relevant tools
buildings. Energy Build 2017;139:476–86. for real-time indoor air quality indicators in low energy buildings. In: Proceedings
[35] Nikdel L, Janoyan K, Bird SD, Powers SE. Multiple perspectives of the value of of the healthy buildings europe 2015 conference; 2015.
occupancy-based HVAC control systems. Build Environ 2018;129:15–25. [65] Jung C-C, Wu P-C, Tseng C-H, Su H-J. Indoor air quality varies with ventilation
[36] Hong T, Koo C, Kim J, Lee M, Jeong K. A review on sustainable construction types and working areas in hospitals. Build Environ 2015;85:190–5.
management strategies for monitoring, diagnosing, and retrofitting the building’s

15

You might also like