Energy and Buildings 117 (2016) 362–371

Contents lists available at ScienceDirect

Energy and Buildings

journal homepage: www.elsevier.com/locate/enbuild

Occupant-oriented mixed-mode EnergyPlus predictive control

simulation
Jie Zhao a,∗ , Khee Poh Lam a , B. Erik Ydstie b , Vivian Loftness a
a
Center for Building Performance and Diagnostics, School of Architecture, Carnegie Mellon University, Pittsburgh, PA, USA
b
Department of Chemical Engineering, College of Engineering, Carnegie Mellon University, Pittsburgh, PA, USA

a r t i c l e i n f o a b s t r a c t

Article history: Model predictive control (MPC) has been studied in the building science realm for about three decades.
Received 9 May 2015 However, the following two aspects of the building control have not been studied thoroughly in MPC
Received in revised form 22 August 2015 research. One is the impact of the mixed-mode cooling system on the active heating ventilation and air
Accepted 13 September 2015
conditioning (HVAC) energy consumption, and the other is the differences of individual thermal comfort
Available online 15 September 2015
preference and its impact on energy. This paper proposes an occupant-oriented mixed-mode EnergyPlus
predictive control system to optimize HVAC energy consumption while meeting the individual thermal
Keywords:
comfort preference. A web-based dashboard is implemented in the test-bed building for three months to
Individual thermal comfort
Mixed-mode
collect individual thermal comfort preference data. The data analysis results suggest that occupants have
Model predictive control various tolerances and preferences about thermal comfort. The simulation results show that, during
EnergyPlus one week of a typical swing season, the mixed-mode system further reduces the active HVAC energy
Energy optimization consumption, and the diversified occupant thermal comfort preference has significant impact on HVAC
Matlab/Simulink energy consumption.
© 2015 Elsevier B.V. All rights reserved.

1. Introduction has been investigated for the past three decades. Generally, MPC
algorithms allow current control inputs to be optimized in real-
Building energy efficiency has been improved greatly for the time control operations with certain constraints while anticipating
past decades, thanks to the adoptions of new technologies and future events using mathematical models. Heating, ventilation, and
more stringent building codes and standards. These energy effi- air conditioning (HVAC) systems can usually take advantage of this
ciency measures mostly happen during the design, construction algorithm to optimize its energy consumption with the constraints
and retrofit phases, which are critical yet very short periods of the of occupant thermal comfort among others. Numerous studies have
entire building life cycle. More light should be shed on building established simulation and experimental results to demonstrate
controls during the operation stage, which has much longer time MPC’s effectiveness in reducing energy and ensuring thermal com-
spent and stronger impact on energy. On the one hand, without fort, such as [1–10].
proper control and management, a building that is designed with Most of MPC studies focus on active HVAC systems, such as
advanced energy-efficient technology may still consume tons of variable air volume (VAV) and radiant heating systems. Mixed-
energy. On the other hand, from the facility managers’ standpoint, mode passive/active HVAC systems that incorporate window
to ensure occupant comfort (without complaints) is often the pri- opening strategies may have a greater energy saving poten-
oritized responsibility. A building control system is the means to tial during cooling and swing seasons in some climates. A few
deliver this commitment. Therefore, for both energy efficiency and researchers have applied MPC algorithms to mixed-mode HVAC
occupant comfort purposes, building control is the key for the daily systems. May-Ostendorp et al. developed an MPC control system
building operation in commercial buildings. that can optimize window opening schedules for a mixed-mode
To achieve the goal of reducing energy and improve occupant HVAC system to reduce energy consumption [11]. On the basis
comfort in commercial buildings, model predictive control (MPC) of this offline optimization framework, Corbin et al. developed
a real-time optimization framework for the MPC system by
linking EnergyPlus, Matlab and a building automation system
∗ Corresponding author at: 5000 Forbes Avenue, MMCH-410-IW, Pittsburgh, PA (BAS). The test cases reported that the MPC system resulted
15213, USA. in “54% energy savings with often improved occupant comfort”
E-mail address: [email protected] (J. Zhao). [12].

http://dx.doi.org/10.1016/j.enbuild.2015.09.027
0378-7788/© 2015 Elsevier B.V. All rights reserved.
 J. Zhao et al. / Energy and Buildings 117 (2016) 362–371 363

Although most of MPC studies on HVAC systems have addressed

thermal comfort, air temperature is often used as the only indi-
cator or constraint. Air temperature is indeed a very influential
factor, but is not the only factor that could influence human ther-
mal comfort. Predicted mean vote (PMV) which can be calculated
by six major human thermal comfort influencers – air temper-
ature, relative humidity, mean radiant temperature, air speed,
metabolic rate, and clothing insulation factors, may better rep-
resent occupant thermal comfort as suggested in the following
studies [5,13–15].
One difference between using PMV and using air temperature as
the thermal comfort indicator is that PMV is able to take into con-
sideration of individual thermal preferences. Several studies have
indicated that occupants do not have uniformed thermal sensa-
tion in the same thermally controlled space [16,17]. In order to
meet the thermal comfort requirements of different occupants,
several studies have developed individual thermal comfort voting
systems to feedback the subjective thermal preference information
into HVAC control systems [14,15,17,18]. For example, Choi [19]
developed a wearable control system that can collect thermal mea-
surement data from individual bio-sensors to automatically adjust
HVAC system setpoints. The experiment and simulation results of
these previous studies show that providing occupants with indi-
vidual thermal controls not only can improve occupant satisfaction
[15,19] and comfort [14,17–19], but also impact the HVAC energy
consumption [14,15,17–19].
This study develops four control schemes using a test-bed office
building to optimize its energy performance while maintaining the
occupant thermal comfort. In Section 2, a systematic method of
creating four control models is presented. First, a baseline rule-
based control model is created to demonstrate a co-simulation
platform and establish a baseline energy and thermal comfort per-
formance. Second, an EnergyPlus predictive control (E+ PC) system
is developed to reduce the active HVAC energy consumption while
maintaining the average occupant thermal comfort. Then, on the
basis of E+ PC, a mixed-mode EnergyPlus predictive control (ME+ PC)
is modeled to further reduce the test-bed building’s energy con-
sumption by incorporating passive cooling strategies in the active
MPC system. At last, in order to meet individual thermal com-
fort requirements, a subjective feedback data collection dashboard
is designed and experimented in the test-bed building for three
month. Based on the data analysis results, an occupant-oriented
mixed-mode EnergyPlus predictive control (OME+ PC) model is
developed and simulated. In Section 3, the simulation results as
well as the data analysis results of the occupant subjective thermal
comfort experiment are presented and discussed. The conclusions
are made at the end of this study.
Fig. 1. Window opening control logic.

2. Methodology
and implementation schemes of the baseline and E+ PC have been
2.1. Case study overview captured in a previous study at [20].

A 2-story, 2262 m2 office building (Phipps Center for Sustain-

able Landscapes (CSL) located in the suburban area of Pittsburgh, 2.2. Mixed-mode EnergyPlus predictive control (ME+ PC)
Pennsylvania, USA) is chosen as a test-bed building to conduct the
simulation and experiment. On the basis of E+ PC, a mixed-mode EnergyPlus Predictive Con-
A baseline control system is developed to establish a co- trol (ME+ PC) model is developed. The objective of ME+ PC is to
simulation framework using Matlab/Simulink and EnergyPlus. The incorporate the passive cooling strategy into E+ PC, so that the
baseline energy and thermal comfort performance using rule-based control system can handle both active HVAC system and passive
algorithm is simulated as a benchmark for the MPC algorithms. On motorized window opening system to further reduce the HVAC
the basis of this baseline model, an EnergyPlus Predictive Control energy consumption with the constraint of occupant thermal com-
(E+ PC) model is developed. The objective of E+ PC is to maintain fort. The motorized window opening control logic is shown in
the same PMV comfort criteria as the baseline control but to min- Fig. 1. The logic is selected based on the actual BAS settings at the
imize the active HVAC power consumption by optimizing the air test-bed building and adjusted for the module of “AvailabilityMan-
handling unit (AHU) supply air temperature. Detailed development ager:HybridVentilation” in EnergyPlus, see details at [21].
364 J. Zhao et al. / Energy and Buildings 117 (2016) 362–371

The content of this part of the dashboard is irrelevant to this

study.

2.4. Occupant-oriented ME+ PC (OME+ PC)

The design of OME+ PC aims to take in the real-time individ-

ual occupant subjective feedback information from the web-based
databases and customize each zone setpoint based on this feedback
information.
Fig. 3 shows the implementation schema of OME+ PC. There
are four layers in this scheme. In the “Data input layer”, weather,
occupancy, lighting, and equipment information is contained in
independent text files (CSV, Comma Separated Values). It is essen-
tial and beneficial to override the EnergyPlus internal weather and
schedule data by using the “Data input layer” and “Receding hori-
zon” module in the “Control layer”. More detailed information is
available from [20]. In the “Data input layer”, a web-based MySQL
Fig. 2. Web-based occupant subjective feedback data collection system. database, which stores the real-time occupant subjective feedback
data, is connected to the “Control layer”.
Two Python [23] scripts are written to query the latest thermal
2.3. Occupant subjective feedback experiment comfort and clothing insulation feedback data from the web-based
MySQL database, respectively. The Python scripts compare the date
E+ PC and ME+ PC are designed to be automated systems that and time of the two pieces of feedback information to ensure they
can optimize energy consumptions of HVAC systems based on are collected at the same time (within the same hour). If the occu-
averaged occupant thermal comfort criteria (average PMV value). pant does not provide both data points at the same time, the Python
Clearly, occupant individual thermal comfort preference is not script assumes the missing data as default value (PMV = 0 and
considered in these systems. It is worthwhile to investigating clo = 1). A Matlab script is written to execute the Python scripts
the actual thermal comfort preference from individual occupants, in the Matlab/Simulink environment.
and then to design a control system to meet the needs of the The “Multi-zone temperature setpoints rule-based control”
preference. Here, a web-based data collection system is devel- module takes in the occupant subjective feedback information and
oped for the occupants in the test-bed building, as shown in computes the adjusted PMV setpoints for the zones that are occu-
Fig. 2. A web-based dashboard is designed to collect occupants’ pied by voters.
individual thermal comfort level with the question of “How are The “Data exchange layer” is used for data communication and
you feeling (thermally)?” and clothing insulation factor with the time synchronization functions, technical details can be found in
question of “What are you wearing?” The thermal comfort band [24]. In the “Simulation layer”, the direct inputs of the EnergyPlus
ranges from −2 to 2 based on the PMV definition. The cloth- model are “AHU supply air temperature setpoint”, “Weather and
ing insulation band ranges from 0.4 to 1.9, representing from schedules information”, and “Multi-zone temperature setpoint”.
“shorts & T-shirts” to “jacket and overcoat”. The occupants are The window opening logic, as shown in Fig. 1, is implemented in
required to provide the two pieces of information concurrently the EnergyPlus model in this layer. The outputs are “HVAC power
when they are not feeling thermally comfortable. All the data demand”, “Multi-zone air temperature, relative humidity, and radi-
are stored in a web-based MySQL [22] database with times- ant temperature”, and “Real simulation time”, which is used for
tamps. time synchronization between Matlab/Simulink and EnergyPlus.

Fig. 3. OME+ PC implementation scheme.

 J. Zhao et al. / Energy and Buildings 117 (2016) 362–371 365

Fig. 4. Outdoor temperature and control variables of E+ PC and ME+ PC in the swing season.

3. Results and discussion baseline control uses typical rule-based control algorithm with
the same model inputs as E+ PC. The mixed-mode baseline control
3.1. Simulation results of ME+ PC is created based on the active mode baseline control and uses the
same window opening logic as ME+ PC.
E+ PC and ME+ PC are simulated for one typical design week in the The first observation is that the mixed-mode baseline control
swing season (October 1st–October 7th) using 2013 annual mete- and ME+ PC can reduce energy significantly during most of the unoc-
orological year (AMY) weather information in Pittsburgh, PA, USA cupied night time, due to the shutdown of AHU while opening
[25]. The same initial warm-up condition is set to be October 1st windows. Another important observation is that ME+ PC has more
for both E+ PC and ME+ PC. power spikes during daytime, compared to E+ PC, especially when
Fig. 4 shows the supply air temperature, outdoor air tempera- transiting from the passive mode to the active mode. The reason
ture, and supply air mass flow rate of E+ PC and ME+ PC. The outdoor could be that the E+ PC’s optimization algorithm shifts the energy
air temperature fluctuates from 10 ◦ C up to 28 ◦ C during the seven consumptions earlier in the morning and maintains the zone air
simulation days, which represents a typical swing season tem- temperature at a lower level, so that during the warmer daytime,
perature range in Pittsburgh, PA. E+ PC uses full air conditioning AHU does not need to consume high power to deliver super cooled
maintains the supply air temperature between 15 and 18 ◦ C and air at a high air flow rate. However, ME+ PC cannot do this because
mass flow rate fairly constant (less than 2 kg/s) for both occupied the passive strategy often takes over the control during unoccupied
and unoccupied periods, thanks to the optimization. During unoc- night time hours as long as the criteria are met.
cupied night time, ME+ PC’s passive mode is active and the AHU Fig. 6 shows the total HVAC energy consumption of the active
supply air flow rate is zero. During occupied daytime, ME+ PC has mode baseline control, mixed-mode baseline control, E+ PC, and
slightly lower or similar supply air temperature and higher flow ME+ PC during the simulation period. There are four pairs of com-
rate for most times with spikes when outdoor temperature rises parisons in this figure.
above 25 ◦ C (e.g., on the afternoon of 10/02 and the morning of Comparison A shows that merely using the passive window
10/06). The results suggest that ME+ PC can reduce fan power and opening strategy on a baseline rule-based control system without
cooling power consumptions greatly by taking advantage of night any predictive control algorithm results in a 33.84% energy sav-
time passive cooling, but it may create spikes at times when the out- ing during the simulation period. Comparison B suggests that only
door conditions are close to the upper setpoint boundary (25 ◦ C). In optimizing the active HVAC control system can save 61.20% energy
future work, window opening logic setpoints could be optimized to during the simulation period. Comparison C shows that compared
identify the optimal balance point between the night time cooling to the rule-based mixed-mode control system, ME+ PC can save
and daytime AHU operation. 44.03% energy during the simulation period, thanks to the pre-
In order to distinguish whether the energy savings are from dictive control algorithm. Comparison D indicates a 4.56% energy
the passive strategy or from the optimization algorithm, the HVAC savings between ME+ PC and E+ PC, which suggests that passive
power demand are compared among the four control models: strategy can be used along with the predictive control algorithm
active mode baseline control, mixed-mode baseline control, E+ PC to further reduce the energy consumption. In summary, the four
and ME+ PC, as shown in Fig. 5. The setting of the active mode comparisons suggest that both passive window opening strategy

Fig. 5. One week plot of the power consumptions of the active mode baseline, mixed-mode baseline, E+ PC, and ME+ PC in the swing season.
366 J. Zhao et al. / Energy and Buildings 117 (2016) 362–371

Fig. 6. One week total HVAC energy consumption of the active mode baseline, mixed-mode baseline, E+ PC, and ME+ PC in the swing season.

and the model-based predictive control algorithm are very effec- (x-axis), respectively. The histograms (N = 60) of the variables are
tive to reduce the active HVAC energy consumption. However, shown on the top and right sides of the plot areas. In general, Sub-
improvements could be made to further investigate and optimize ject A feels cold for the majority of votes regardless of actual air
the passive control strategies in order to make the mixed-mode temperature (20–25 ◦ C) measurements. One noticeable pattern is
predictive control works better when the MPC algorithm has been that all the “very cold” votes are the times when the air temperature
used in the controls. drops below 21 ◦ C. Fig. 9 shows that most of the “feeling cold” votes
Fig. 7 shows the average zone PMV comparison between E+ PC are correlated with clothing insulation factor above 1.5, which sug-
and ME+ PC. In general, E+ PC has better capability to maintain the gests that although Subject A has adapted his/her clothes based on
PMV value under 0.5 during the occupied hours. E+ PC keeps a the thermal comfort feeling, he/she was still feeling cold during the
lower PMV value during unoccupied night hours to “prepare” for study period.
the daytime temperature rise. In contrast, ME+ PC has a few minor Figs. 10 and 11 show Subject B’s marginal plots of air tempera-
violations during warmer hours during the daytime. The comfort ture and clothing insulation (y-axis) vs. the comfort vote (x-axis),
criterion is loosen during weekends for October 6th and 7th. respectively. The histograms (N = 36) of the variables are shown on
the top and right sides of the plot areas. In general, Subject B feels
3.2. Occupant subjective feedback data analysis cold for the majority of votes regardless of actual air temperature
(20–25 ◦ C) measurements. Fig. 11 shows that almost all the “feeling
In order to find useful patterns of occupant thermal com- cold” votes (except for one vote) are correlated with clothing insula-
fort preference, the subjective thermal comfort votes from the tion factor around 0.5, which suggests that Subject B has relatively
web-based database and the objective thermal environment mea- thin clothes in the office compared to the typical business dress.
surement data from the test-bed building BAS system are mapped Arguably, Subject B could have improved his/her thermal comfort
based on spatial and temporal information [21]. A total of 15 volun- level easily by adding a jacket or another layer of clothes. This result
teers have participated and provided their subjective feedback on is very different from the data of Subject A, who has already had
thermal comfort and clothing insulation information from October thick clothes but still feels cold.
to December 2013. Among the 15 volunteers, only three occupants Figs. 12 and 13 show Subject C’s marginal plots of air tempera-
(20%) have more than 30 effective voted instances on both thermal ture and clothing insulation (y-axis) vs. the comfort vote (x-axis),
comfort and clothing insulation concurrently. Eight (53%) occu- respectively. The histograms (N = 41) of the variables are shown
pants have less than 30 effective voted instances and four occupants on the top and right sides of the plot areas. In general, Subject C
(27%) have no effective voted instance over the experiment period. feels warm, especially when the air temperature is above 22 ◦ C.
From the statistics point of view, three datasets with more than Fig. 13 shows the clothing insulation factor is mostly below 1.0,
30 effective voted instances are mapped with the measured air which suggests that Subject C has relatively thin clothes in the office
temperature in their corresponding thermal zones. compared to typical business dress, but the subject still feels warm
Figs. 8 and 9 illustrate Subject A’s marginal plots of air temper- in general. This is clearly different from the other two subjects’
ature and clothing insulation (y-axis) vs. the thermal comfort vote profiles.

Fig. 7. Thermal comfort performance of E+ PC and ME+ PC in the swing season.

 J. Zhao et al. / Energy and Buildings 117 (2016) 362–371 367

Fig. 8. Marginal plot of air temperature vs. comfort vote of Subject A.

Fig. 9. Marginal plot of clothing insulation vs. comfort vote of Subject A.

Fig. 10. Marginal plot of air temperature vs. comfort vote of Subject B.
368 J. Zhao et al. / Energy and Buildings 117 (2016) 362–371

Fig. 11. Marginal plot of clothing insulation vs. comfort vote of Subject B.

Fig. 12. Marginal plot of air temperature vs. comfort vote of Subject C.

Fig. 13. Marginal plot of clothing insulation vs. comfort vote of Subject C.
 J. Zhao et al. / Energy and Buildings 117 (2016) 362–371 369

Table 1 ME+ PC, as illustrated in Fig. 16. During the one week simulation
Thermal comfort and clothing insulation preferences in OME+ PC.
period in the swing season, OME+ PC has a 37.37% energy reduction
Thermal comfort Clothing insulation compared to ME+ PC.
preference range in preference (Clo) However, the energy reduction is mainly caused by the
occupied hours (PMV)
“warmer” thermal preference of Subject A and B in the cool-
Default value (−0.5, 0.5) 1.00 ing dominated period, so the setpoints of zone temperatures are
Occupant A (0.19, 1.19) 1.25 higher than the default value. The occupant preference can vary
Occupant B (−0.14, 0.86) 0.57
significantly based on season, location, physical and psychologi-
Occupant C (−0.78, 0.22) 0.74
cal changes. For instance, even with the same group of occupants
given the same setting of OME+ PC, the energy consumption could
be increased during heating seasons, since the occupants demand
3.3. Simulation results of OME+ PC warmer environment than the default mode. Therefore, the conclu-
sion of OME+ PC simulation study is that introducing the individual
OME+ PC is simulated using the same weather, occupancy, occupant thermal comfort preference in the control system may
equipment, and lighting schedule inputs as the ones used in ME+ PC. have significant energy impact while providing customized ther-
The only different input between OME+ PC and ME+ PC is the occu- mal environment to individuals. More in-depth investigation could
pant thermal comfort preference range. As shown in Table 1, ME+ PC be conducted in future.
uses the default value (PMV = 0) for all the occupants in the six open
office zones, while in OME+ PC, three different occupant preferences
learned from the occupant subjective feedback study are simulated 3.4. Discussions
in the three open office zones, respectively. The default value is
implemented in the other three zones. Three limitations need to be discussed in this study. First, one
Fig. 14 shows the simulation results of the PMV values of differ- week of simulation has been conducted for different control models
ent zones in OME+ PC. It is clear that the default PMV performance in various weather conditions. The baseline and E+ PC are simu-
obeys the original (−0.5, 0.5) range in occupied hours, but the three lated in both heating and cooling seasons. The E+ PC, ME+ PC, and
adjusted thermal zone PMV values are shifted either up or down OME+ PC are simulated in swing seasons. For generating one hourly
to meet the individual comfort preference. Therefore, this result of actual control output, the control optimization algorithm needs
demonstrates that OME+ PC is capable of providing individualized about 5 min of simulation time in the Matlab/Simulink and Energy-
thermal environment based on occupants’ comfort preferences. Plus co-simulation platform. Therefore, one week of actual control
In order to further investigate the energy implications of simulation takes about 14 h to finish. Although the computation
allowing occupants to choose their own thermal comfort prefer- time can be improved, in reality, when the control step is adjusted
ences, the OME+ PC’s AHU supply air temperature, flow rate, and hourly, five minutes of optimization is still acceptable. However,
power consumption data are compared with ME+ PC’s. As shown more long-term investigations are needed to discover the annual
in Fig. 15, OME+ PC has generally higher supply air temperature impact of the different control schemes on energy and occupant
and lower mass flow rate during occupied daytime for cooling. comfort.
In addition, OME+ PC has fewer occupied cooling hours during The second limitation is that a rule-based passive cooling control
the first four weekdays. In accordance with the performance of algorithm is implemented together with the MPC control algorithm
control variables, the power consumption of OME+ PC is lower than for the active HVAC system. The control system worked well in

Fig. 14. Thermal comfort performance of different zones in OME+ PC in the swing season.

Fig. 15. Outdoor temperature and control variables of ME+ PC and OME+ PC in the swing season.
370 J. Zhao et al. / Energy and Buildings 117 (2016) 362–371

Fig. 16. One week plot of the power consumptions of ME+ PC and OME+ PC in the swing season.

terms of saving energy and maintaining occupant thermal comfort further improvement of the mixed-mode control algorithm could
in general, but there are times when thermal comfort is com- be made.
promised and there are other times when the active system has Furthermore, the sample size and the climate conditions of
to push its operation to extremes in order to meet the setpoint. the occupant subjective thermal comfort feedback study could
In future, researchers could first focus on improving the passive be enlarged and diversified. This study has already found very
cooling control algorithm with physical experiments and/or sim- interesting patterns, despite the small sample size (15 people)
ulations to identify a proper algorithm for a specific building in a and the relative short duration (three months). One could imagine
specific location. Then based on the optimized passive algorithm, that if future studies could incorporate more individuals under

Fig. 17. Different control schemes, as well as their methods and outcomes.
 J. Zhao et al. / Energy and Buildings 117 (2016) 362–371 371

different seasons and locations that could also represent culturally 1038139) and Phipps Conservatory and Botanical Gardens for fund-
behavioral differences, the findings would be very valuable for ing the study. The authors would also like to acknowledge Bertrand
understanding occupant thermal comfort and behavior, and for Lasternas, Ray Yun, Haopeng Wang, Omer Karaguzel, and Xuan Luo
developing more appropriate occupant-oriented control systems. from Carnegie Mellon University Center for Building Performance
and Diagnostics for their contributions to this study.
4. Conclusions
References
This paper describes the development procedures and sim-
ulation results of an occupant-oriented mixed-mode EnergyPlus [1] M. Castilla, J.D. Alvarez, M. Berenguel, F. Rodriguez, J.L. Guzman, M. Perez, A
comparison of thermal comfort predictive control strategies, Energy Build. 43
predictive control (OME+ PC) system. Three interim control mod- (2011) 2737–2746.
els are created to compare different energy saving and thermal [2] B. Dong, K.P. Lam, Building energy and comfort management through
comfort improving strategies, which are the baseline rule-based occupant behaviour pattern detection based on a large-scale environmental
sensor network, J. Build. Perform. Simul. 4 (4) (2011) 359–369.
control (Baseline), EnergyPlus predictive control for the active [3] D. Gyalistras, Use of Weather and Occupancy Forecasts for Optimal Building
HVAC system (E+ PC), and mixed-mode EnergyPlus predictive con- Climate Control (OptiControl), Swisselectric Research, Zurich, 2010.
trol (ME+ PC). Fig. 17 shows the development processes, methods, [4] G.P. Henze, R.H. Dodier, M. Krarti, Development of a predictive optimal
controller for thermal energy storage systems, HVAC&R Res. 3 (3) (1997)
and outcomes of the four control schemes. 233–264.
The first set of outcomes is that a co-simulation platform [5] Y. Yu, V. Loftness, D. Yu, Multi-structural fast nonlinear model-based
is created for EnergyPlus and Matlab/Simulink via MLE+. The predictive control of a hydronic heating system, Build. Environ. 69 (2013)
131–148.
co-simulation platform is tested numerous times during the
[6] R.Z. Freire, G.H.C. Oliveira, N. Mendes, Predictive controllers for thermal
development process and it can be systematically updated with comfort optimization and energy savings, Energy Build. (2008) 1353–1365.
minimum effort. The second set of outcomes is that during the sim- [7] S. Yuan, P. Ronald, Multiple-zone ventilation and temperature control of a
ulation periods (heating, cooling, and swing seasons), E+ PC, ME+ PC, single-duct VAV system using model predictive control, Energy Build. 38
(2006) 1248–1261.
and OME+ PC can reduce HVAC energy consumptions and main- [8] D. Kolokotsa, A. Pouliezos, G. Stavrakakis, C. Lazos, Predictive control
tain occupant thermal comfort with the model-based predictive techniques for energy and indoor environmental quality management in
control algorithm. The third set of outcomes is that using the mixed- buildings, Build. Environ. 44 (2009) 1850–1863.
[9] B. Paris, J. Eynard, S. Grieu, T. Talbert, M. Polit, Heating control schemes for
mode active/passive control algorithm, more energy savings can be energy management in buildings, Energy Build. 42 (2010) 1908–1917.
achieved and the occupant thermal comfort can be maintained. Last [10] S. Privara, J. Siroky, L. Ferkl, J. Cigler, Model predictive control of a building
but not least, the final outcome of implementing the real-time ther- heating system: the first experience, Energy Build. 43 (2011) 564–572.
[11] P. May-Ostendorp, G.P. Henze, C.D. Corbin, B. Rajagopalan, C. Felsmann,
mal comfort feedback system, OME+ PC can provide individualized Model-predictive control of mixed-mode buildings with rule extraction,
thermal environment for occupants and has considerable impact on Build. Environ. 46 (2011) 428–437.
the HVAC energy consumption. These conclusions are drawn based [12] C.D. Corbin, G.P. Henze, P. May-Ostendorp, A model predictive control
optimization environment for real-time commercial building application, J.
on several one-week simulation results in the study in typical heat- Build. Perform. Simul. 6 (3) (2012) 159–174.
ing, cooling and swing seasons, thus the results cannot be applied to [13] K.H. Yang, C.H. Su, An approach to building energy savings using the PMV
annual-based simulation studies at this point. Future studies could index, Build. Environ. 32 (1) (1997) 25–30.
[14] D. Daum, F. Haldi, N. Morel, A personalized measure of thermal comfort for
run annual simulation studies and investigate the annual energy
building controls, Build. Environ. 46 (1) (2011) 3–11.
and comfort impact of the control schemes. [15] Y. Murakami, M. Terano, K. Mizutani, M. Harada, S. Kuno, Field experiments
In the data analysis study of the occupant subjective thermal on energy consumption and thermal comfort in the office environment
feedback, a total of 11 out of 15 volunteers express their opinions controlled by occupants’ requirements from PC terminal, Build. Environ. 42
(2007) 4022–4027.
on their thermal comforts when they do not feel thermally com- [16] J. Park, Post-occupancy evaluation for energy conservation, superior IEQ &
fortable, which indicates thermal comfort is something important increased occupant satisfaction, in: World Workplace 2013, Ifma, 2013.
for the test group to consider at workplaces. However, people have [17] L. Klein, J.-Y. Kwak, G. Kavulya, F. Jazizadeh, B. Becerik-Gerber, P.
Varakantham, M. Tambe, Coordinating occupant behavior for building energy
very different sensitivities and tolerances to thermal comfort. Three and comfort management using multi-agent systems, Autom. Constr. 22
subjects voted over 30 times during the three months experiment (2012) 525–536.
periods, but most people voted less than 30 times. In addition, [18] Y.S. Lee, A.M. Malkawi, Simulating multiple occupant behaviors in buildings:
an agent-based modeling approach, Energy Build. 69 (2014) 407–416.
among the three subjects, the data show very different comfort [19] J.H. Choi, CoBi: Bio-Sensing Building Mechanical System Controls for
preferences and clothing choices. However, the thermal comfort Sustainably Enhancing Individual Thermal Comfort (Ph.D. Dissertation),
preference and clothing choice of each individual are mostly con- Carnegie Mellon University, Pittsburgh, 2010.
[20] J. Zhao, K.P. Lam, B.E. Ydstie, O.T. Karaguzel, EnergyPlus model-based
sistent for the majority of time during the entire experiment period. predictive control within design–build–operate energy information
Therefore, it is necessary to develop a control system that can modelling infrastructure, J. Build. Perform. Simul. (2014) 1–14.
accommodate individual occupant thermal comfort preference, [21] J. Zhao, Design-Build-Operate Energy Information Modeling for
Occupant-Oriented Predictive Building Control (Dissertation), Carnegie
and based on the individual preference, the control system could
Mellon University, 2015.
make consistent adjustment to meet individual needs. [22] Oracle, MySQL, 2014, http://www.mysql.com/ (accessed 27.04.14).
[23] Python, Python Programming Language – Official Website, 2012, http://
python.org/ (accessed 20.01.12).
Acknowledgments [24] W. Bernal, M. Behl, T. Nghiem, R. Mangharam, MLE+: a tool for integrated
design and deployment of energy efficient building controls, in: 4th ACM
Workshop on Embedded Sensing Systems for Energy-Efficiency in Buildings
The authors would like to acknowledge National Science Foun- (BuildSys ‘12), Toronto, 2012.
dation (NSF) Emerging Frontiers in Research and Innovation (EFRI) [25] DOE, Weather Data for Simulation, 2014, http://apps1.eere.energy.gov/
in Science in Energy and Environmental Design (SEED) (Award #: buildings/energyplus/weatherdata simulation.cfm (accessed 22.08.15).