You are on page 1of 20

59

Vol. 2, No. 1 HVAC&R Research January 1996


Introduction
The most common building zone strategy for cooling is termed night setup control. During occupied
hours, zone conditions are typically controlled at constant set points that maintain acceptable comfort. Dur-
ing unoccupied times, the equipment is turned off and the zone temperature is allowed to float. Night setup
control strategies minimize the effects of building thermal storage. However, in many commercial build-
ings, the building mass represents a significant thermal storage medium that can be used to reduce the oper-
ating costs associated with cooling. For example, an optimal controller might precool a building during the
unoccupied period and control the storage discharge process by varying the set points within acceptable
comfort bounds during occupancy.
A number of simulation studies have shown how significant reductions of operating costs in buildings
are achieved by proper precooling and discharge of building thermal storage. The savings result from both
utility rate incentives (time-of-use and demand charges) and improvements in operating efficiency due to
night ventilation cooling and improved chiller performance (lower ambient temperatures and more even
loading). Braun (1990) showed significant energy cost savings (10% to 50%) and peak power reductions
(10% to 35%) over night setup control in a comprehensive simulation study which considered several
building types and weather conditions. The percent savings were found to be most significant when low
ambient temperatures allowed night ventilation cooling to be performed. Andresen and Brandemuehl
(1992) simulated a typical zone of an office building and reported reductions in peak cooling loads of 10%
to 50% depending on the control strategy which was used. Other simulation studies which yielded compa-
A Simplified Method for Determining Optimal Cooling
Control Strategies for Thermal Storage in Building
Mass
Kevin Keeney
Student Member ASHRAE
James Braun, Ph.D., P.E.
Member ASHRAE
Previous simulations and experiments have shown that significant operating savings can be realized when
cooling commercial buildings if the building structure is used for thermal storage. Research has also shown
that the control strategy must be tailored to the application to achieve these savings while maintaining occu-
pant comfort. Developing cooling control strategies which utilize the thermal mass of a building is a formi-
dable optimization problem. By examining optimal cooling results covering a wide range of buildings,
cooling plants, weather, utility rates, and internal gains, this study has reduced the dimensionality of the
optimal control problem. Two simplified approaches were developed which each employ two control vari-
ables while the building is unoccupied, in conjunction with a set of rules for the occupied period. These rules
were expressed in terms of occupant comfort. The simplified strategy achieved 95% and 97%, respectively,
of the optimal savings relative to conventional control. Using hourly time-steps, the daily optimization prob-
lem was reduced from 24 to 2 variables. In addition to reducing the computational requirements required to
study optimal control and develop building specific control strategies, these simplifications could be used in
the development of an on-line controller.
Kevin Keeney is a graduate student and James E. Braun is a professor in the School of Mechanical Engineering, Purdue University,
West Lafayette, Indiana.
Table of Contents
60 HVAC&R RESEARCH
rable results include Rabl and Norford (1991), Snyder and Newell (1990), and Golneshan and Yaghoubi
(1990).
These simulation studies demonstrated that the savings potential and best control strategy are both
very dependent on the system as well as on particular weather conditions. Improper precooling can actually
result in costs that are greater than those associated with conventional control. The importance of optimiz-
ing the control for each application has also been demonstrated by the following experimental evaluations.
Conniff (1991) used a test facility at the National Institute of Standards and Technology (NIST) to study
the use of building thermal mass to shift cooling load. The facility was designed to represent a zone in a
typical commercial office building. Several control strategies were considered in these tests. No attempt
was made to optimize the control strategies for this facility. The most effective strategy tested for peak
reduction did not utilize precooling but, instead, used a constant zone temperature for the first seven hours
of occupancy, followed by a limit on the amount of cooling supplied to the zone. This strategy lowered the
peak cooling demand by up to 15% of the lighting power when compared to night setup control. Other
strategies which utilized precooling resulted in minimal cooling demand reductions (3%). Since thermal
comfort was not evaluated during the tests, additional precooling might have been possible without sacri-
ficing occupant comfort.
Ruud et al. (1990) performed two experiments on an office building located in Jacksonville, FL. The
control strategy called for cooling with a supply air temperature of 50F from 5 P.M. until 5 A.M. The first
experiment resulted in several occupant comfort complaints which were addressed by adding a warmup
period just prior to building occupancy. No comfort complaints were reported after the warmup period was
initiated. The results showed 18% of the total daytime cooling load had been shifted to the night period
with no reduction in peak demand. Again, the control strategies used in this study were not optimized for
the Jacksonville building.
More recently, Morris et al. (1994) devised and performed a set of experiments at the same facility used
by Conniff (1991) in order to validate the potential for load shifting and peak reduction associated with
optimal control. Prior to performing experiments, a simulation of the test facility was developed that
included detailed models of the structure, cooling system, and comfort of a human occupant. Optimization
techniques were applied to the simulation model in order to determine control strategies used in the two
separate tests. The first control strategy was designed to minimize total energy costs and resulted in the
shifting of 51% of the total cooling load to the off peak hours. The second control strategy was designed to
minimize the peak electrical demand and resulted in a 40% reduction in peak cooling load. In both of these
tests, data were collected to measure the occupant thermal comfort. Thermal comfort was maintained
within acceptable limits throughout both of the experiments. The results of Morris et al. (1994) were more
encouraging than Conniff (1991) for the same test facility because the control was optimized. Another
important result of this work was the validation of the model used to develop the optimized control strate-
gies. The time distribution of cooling load from the building model closely matched the experimental
results under both night setup and optimal control.
The simulation and experimental studies cited above demonstrated the significant potential for cost sav-
ings through proper control of building thermal storage and documented the need for optimizing the con-
trol. The next step in the research process was to develop practical methods for controlling building
thermal storage that could be readily implemented in building control systems and give near-optimal per-
formance. However, this is an extremely difficult problem and no general control strategies have been pre-
sented and compared with optimal control benchmarks in the literature.
This paper presents a simplified simulation-based method for determining the best zone setpoints that
minimize daily energy costs for any system over a 24-hour period. Calculation of the energy costs consid-
ered the effects of a variable energy cost over time, but did not include demand charges. This simplified
method is compared with a more detailed benchmark approach based upon the work of Morris et al. (1994)
that involves the optimization of 24 variables for each zone (the zone temperatures for each hour of the
day). The simplified method presented here reduces this problem to the determination of only two vari-
ables. This result is significant for two reasons:
1. The simplified optimization procedure can be implemented very easily with low computational require-
ments to provide a global optimum for any application. Other researchers can now easily develop tools
VOLUME 2, NUMBER 1, JANUARY 1996 61
useful in the development and evaluation of generic control strategies, or ones that could be used
directly to develop site-specific control strategies.
2. The simplified optimization method could itself be the basis of an on-line, adaptive algorithm for con-
trolling building thermal storage. In order to pursue this approach, it would be necessary to develop
simplified models for building dynamics and cooling plant performance that could be easily learned
with a minimum number of sensors.
Approach
Figure 1 shows a block diagram of the approach taken in the development of the simplified optimal con-
trol strategies described in this paper. The first step involved developing a model which considered all of
the parameters that influence the control of building thermal mass. The following parameters were identi-
fied as factors for consideration in this study:
1. Building construction
2. Utility rate structure
3. Weather
4. Internal gains
5. Cooling plant
Simplified control strategies were developed that were compared to optimal and night setup control over
a wide range of parameter values. Control strategies were evaluated on the basis of the total daily cooling
cost and the occupant comfort. Night setup control was selected as the conventional control baseline for
comparison. Under night setup control, the zone temperature is fixed while the building is occupied. When
the building is unoccupied, the set point is raised to a high enough value to prevent the cooling plant from
operating.
The optimal control strategy is defined as the strategy which results in the minimum cooling cost with-
out violating comfort conditions while the building is occupied. Together, the night setup cost and optimal
control cost represent the high and low limits, respectively, for the cost under an acceptable alternative con-
trol strategy.
Simulation Tool
The simulation tool used to develop and evaluate control simplifications and strategies presented in this
paper is based on the model used and validated by Morris et al. (1994). Significant modifications include
the addition of a zone humidity model, improved optimization procedures, and the ability to specify zone
set points in terms of an occupant comfort index. A detailed description of the simulation tool can be found
in Keeney and Braun (1995).
Figure 1. Approach to developing optimal control simplifications
62 HVAC&R RESEARCH
Figure 2 shows a block diagram representation of the simulation tool. The controller generates the zone
set points for the building model with respect to each control strategy under consideration. The building
model uses the control inputs with ambient conditions from the weather model to calculate the total cooling
requirements (sensible and latent). Zone cooling loads are then used by the plant model to calculate the
cooling power required to maintain the desired zone conditions. The comfort model uses the zone condi-
tions calculated in the building model to calculate an occupant comfort index based on the Fanger predicted
mean vote (PMV) scale (Fanger 1970). The various component models are described in more detail below.
Controller
The controller generates the zone set points for the building model. Each hourly set point may be
expressed in terms of a zone temperature set point, a sensible cooling applied to the zone, or a set point
value for the occupant comfort index. Under night setup control, the nighttime zone temperature set point is
set to a high value so that all cooling plant equipment stays off at night and the zone temperature floats.
At occupancy, the zone temperature set point is set to the highest fixed value within acceptable limits that
maintains occupant comfort throughout the occupied period.
When acting as an optimal controller, the controller uses numerical optimization to minimize the energy
costs for cooling for one day. The cost function is given by:
(1)
where is the vector of zone set points, k is the time stage, N is the number of time stages in a day,
is the total coil load, t is the simulation time-step, COP is the cooling plant coefficient of performance,
is the fan input power, and R
k
is the utility rate multiplier defined as the cost of electricity during the
on-peak period divided by the off-peak cost.
At each time-step, the following constraint equations must be satisfied:
(2)
Figure 2. Simulation tool block diagram
f u ( )
Q

coil k ,
t
COP
k
------------------------ P
fan k ,

t +


R
k
k 1 =
N

=
u
Q

coil
P

fan
Q

coil k ,
Q

cap

VOLUME 2, NUMBER 1, JANUARY 1996 63


(3)
(4)
where is the design load on the cooling plant, T
z
is the zone air temperature, T
s
is the temperature of
the air being supplied to the zone by the cooling plant, and PMV
min
and PMV
max
are the minimum and
maximum acceptable values for the predicted mean vote comfort index. The upper and lower PMV limits
were set to 0.5 and 0.5, respectively, during occupied periods and were not enforced during unoccupied
periods.
The zone temperature set point was limited to the temperature of the zone supply air exiting the cooling
coil [55F (12.8C)]. This prevented the simulation tool from attempting to cool the space to an unrealisti-
cally low temperature. Equation (2) limits the coil load to the capacity of the cooling plant.
The constraint of Equation (4) was used to ensure that comfort was maintained while the building was
occupied. An exponential penalty function was used to implement the comfort constraint. The penalty
function added an artificial cost when the zone PMV during the occupied period was outside of the PMV
limits and is given by:
(5)
where Pen
pmv
is the penalty associated with the violation, A is the penalty step constant, and B is the pen-
alty rate constant. The initial additional step cost is set by the variable A. As the violation increases, the
penalty cost increases exponentially at a rate set by B. To ensure that the optimal solution did not result in a
comfort violation, A was set to ten times the maximum hourly cost for cooling on the design day, and B was
chosen so that the penalty was doubled when the zone PMV violation was 0.1.
The penalty is added directly to the cost function given in Equation (1) to form the cost function used by
the optimization routine:
(6)
The complex method (Rao 1984) was used to minimize the cost function defined by Equation (6) with
respect to . This is a direct search method that generates a shape in the control variable space that always
encloses the minimum point. The shape is composed of twice as many points as there are control variables
and is termed a complex. The points on the complex are contracted or expanded based on updates from cost
function evaluations until the convergence tolerance or the limit of maximum function evaluations is
reached.
Plant Power Model
Figure 3 shows a schematic of the mechanical cooling system considered in this project. The plant pro-
vides conditioned air through an air handling unit which cools and dehumidifies the air. The air handler
T
z k ,
T
s

PMV
min k ,
PMV
k
PMV
max k ,

Q

cap
Pen
pmv
A B PMV PMV
limit
( ) exp =
f u ( )
Q

coi l k ,
t
COP
k
------------------------ P
fan k ,

t +


R
k
Pen
pmv k ,
+
k 1 =
N

=
u
64 HVAC&R RESEARCH
uses chilled water which is provided by a vapor compression refrigeration unit. The condenser is cooled
with water from a cooling tower.
Empirical functions developed by Braun (1990) were used to model the power consumption of the
mechanical cooling plant. The model determines the overall cooling plant coefficient of performance
(COP) as a function of the plant part-load ratio and the ambient air wet bulb temperature. The plant part-
load ratio is defined as the ratio of the cooling load to the design load of the cooling plant. This COP
includes all of the cooling plant equipment (chillers, pumps, etc.) with the exception of the air handler fans.
The correlations were developed from detailed system simulations over a wide range of operating condi-
tions.
Variable speed and outlet damper fan power models were considered in this project. The fan power
models are described in Knebel (1983). The fan power was calculated using a fan part-load fraction of 0.2
for flow-rates less than 20% of the design flowrate.
Cooling Load Model
The overall cost function given by Equation (6) is a function of the total cooling coil load. The coil load
can be expressed as:
(7)
where is the sensible zone cooling load, is the latent zone cooling load, and is the
cooling load due to ventilation.
The multi-zone building energy analysis subroutine (Type 56) of the dynamic simulation program TRN-
SYS (Klein et al. 1990) was used in this study to model the thermal behavior of a user-specified multi-zone
building. The building model uses an overall energy balance to estimate the net sensible energy gain to the
zone air. Transient conduction in the multi-layered walls is modeled with the transfer function approach of
Stephenson and Mitalis (1971). The transfer functions calculate the heat flux through the walls using a time
history of air temperatures and surface heat fluxes. In this application, the Type 56 subroutine calculated
the sensible cooling required to achieve a desired zone air temperature set point or the zone temper-
ature for a specified sensible cooling rate. The building model also calculated the air temperature, surface
Figure 3. Mechanical cooling system
Q

coil
Q

coil
Q

s z ,
Q

l z ,
Q

vent
+ + =
Q

s z ,
Q

l z ,
Q

vent
Q

s z ,
VOLUME 2, NUMBER 1, JANUARY 1996 65
temperatures, and humidity ratio for the zone. These parameters were collectively termed the zone condi-
tions and their values were used by the comfort model to estimate the occupant comfort within the space.
One form of precooling is night ventilation, where cool night air is brought into the zone by a fan to pro-
vide sensible cooling. Depending on the moisture content of the air, it may also add unwanted moisture to
the space which will be realized as a future latent cooling load when the mechanical cooling plant is oper-
ated. To adequately compare alternative control methods, moisture effects must be accounted for in both
the building model and the cooling coil model. In the building model, the zone humidity was calculated
using a single-node lumped capacitance moisture balance. To model the latent load of the coil, an empirical
model with coefficients developed by Braun (1990) was used. The coil sensible heat ratio was correlated as
a function of the coil sensible load, inlet air dew-point temperature, and outlet air dry bulb temperature.
A dry bulb economizer was included in the cooling plant model. The economizer cycle is often used in
commercial buildings when cooling is required and the outdoor air temperature is lower than the zone air
temperature set point. In this situation, the outdoor air can be used in place of return air to lower the sensi-
ble cooling load of the coil. In some cases, the air may be introduced to the space without any additional
cooling. This means of cooling is often called free cooling even though fan energy is consumed and must
be considered in the overall system analysis. In the economizer model, the inlet air to the coil was assumed
to be at ambient conditions when the outdoor air temperature was lower than the return air from the zone.
When the building was occupied, a minimum fresh air requirement of 20 cfm (9.4 L/s) per occupant
was in effect. The minimum fresh air requirements were based on guidelines for office spaces (ASHRAE
1989b). When the building was occupied and the ambient temperature was greater than the zone tempera-
ture, the minimum flowrate of fresh air was mixed with the return air prior to the cooling coil.
Comfort Model
In this study, comfort indices were calculated from the correlations developed by Fanger (1970). The
Fanger comfort model provides a comfort index termed the Predicted Mean Vote (PMV) that represents the
comfort of an average building occupant. A PMV value of zero is defined as neutral comfort (neither cool
or warm), positive PMV values indicate a warm environment, and negative values indicate a cool environ-
ment. For this study, the PMV value during occupied periods was restricted to values between 0.5 as rec-
ommended by ASHRAE (1989a).
The inputs to the comfort model are air temperature, water vapor pressure, mean radiant temperature,
occupant activity level, and occupant clothing. The zone vapor pressure was calculated from the zone
humidity ratio assuming both the water vapor and the dry air act as ideal gases. The mean radiant tempera-
ture was calculated from the wall surface temperatures using angle factor algorithms described in Fanger
(1970). The occupant activity level was assumed to be constant at a value of 60.0 kcal/m
2
h (70 W/m
2
).
This corresponds to light office activity. Clothing values were chosen to represent typical office attire and
were also assumed to be constant throughout each day.
When the building control is expressed as a zone PMV set point, the zone temperature set points are
adjusted to produce a desired PMV value for the space. Adjusting the zone air temperature set point also
influences the latent coil load and consequently the zone air humidity ratio. Both the zone air temperature
and the zone humidity ratio are variables in the comfort model used in this study. To promote stability in
the simulations, the zone humidity from the previous time-step was used in the calculation of PMV when
PMV index control was used.
Weather Model
Weather data in the form of ambient temperature, humidity, and solar radiation values were also needed
for the simulations. The simulations required weather data which could be used to compare different strate-
gies over a reproducible set of ambient conditions. The ambient air state data were calculated using an
algorithm developed by Erbs (1984) that generates statistical hourly temperature and humidity profiles
from an input daily average clearness index and daily average temperature. The clearness index k
t
is the
ratio of the horizontal surface solar radiation on the ground to the horizontal extraterrestrial radiation.
Solar radiation values were needed for all exterior surfaces. The hourly horizontal surface radiation was
modeled using additional correlations by Erbs (1984). Solar radiation levels were calculated for all surfaces
using the hourly horizontal radiation values with correlations given in Duffie and Beckman (1991). In this
66 HVAC&R RESEARCH
study, a range of daily weather profiles were generated by varying the average outdoor temperature and the
clearness index.
System Simulations
The component model subroutines were linked with a main program to form the simulation tool. The
main program was used to load data for the simulation, call the component models and write the output
data to the output file.
Daily simulations were used to test the simplified optimal control strategies developed in this work. At
each time-step, the system of nonlinear equations resulting from the component models were solved. Given
the number of calculations needed for optimization, it was appropriate that the time-step should be chosen
to be as large as possible while capturing the dynamics of the building. The time constants associated with
the heat transfer processes in buildings are in the order of several hours. Morris et al. (1994) showed that
the dynamics of precooling could be accurately modeled using one hour time-steps. For these reasons,
hourly time-steps were used in this project.
The initial conditions of the building affect the results of the simulation. The building model subroutine
allows the initial temperatures of the building mass to be set to a single uniform temperature, but a uniform
temperature distribution is not likely to occur in a building under normal operation. To provide more realis-
tic initial conditions and allow the costs between different control strategies to be compared, the daily sim-
ulations were run until steady-periodic conditions were achieved. Figure 4 shows the cooling load profile
for a zone using night setup control over a 7 day period. The wall temperatures were uniform at the start of
the simulation. The conditions at the end of each day were used as the initial conditions for the following
day.
The cooling load profile shown in Figure 4 changes significantly during the first 3 days of the simulation
and then approaches a steady-periodic behavior. Steady-periodic conditions were assumed to occur when
the daily total zone sensible cooling changed less than 0.5% between simulation days. Figure 4 clearly
demonstrates the thermal storage potential of the building mass.
Figure 4. Effect of initial conditions on zone cooling requirements
VOLUME 2, NUMBER 1, JANUARY 1996 67
A summary of the parameters used in this study is given in Table 1. The parameters were chosen such
that a wide range of cooling applications were considered. The descriptors for the building constructions
and cooling plants are given below.
Four different types of building zones were considered in this study, as described in this section. A
detailed descriptive file was written for each of these zones in a format used by TRNSYS subroutines. The
input files are listed in Keeney and Braun (1995). The different building models represented heavy and
light constructions and varying degrees of ambient coupling.
The first building considered was a single-story block wall construction building. This building was
designed to represent a small, low-rise office building which is typically used for professional offices and
small retail businesses. Exterior walls were constructed of brick and insulated with extruded polystyrene
wallboard. A cross-section of the exterior wall construction is shown in Figure 5. Interior walls were con-
structed of lightweight aggregate block with plaster finish surfaces. The foundation was modeled as slab on
grade construction. The roof was modeled as a built-up roof layer placed over a lightweight concrete slab
with a suspended ceiling. The building was rectangular in shape with dimensions of 100 ft by 75 ft (30.5 m
Table 1. Parameters Studied for Optimal Control Simplification
Parameter Values Used
Building construction Internal, block, frame, curtain
Utility rate multiplier 1.0, 1.5, 2.0
Clearness index 0.45, 0.60, 0.75
Average ambient temperature 65, 70, 75, 80F (18, 21, 24, 27C)
Internal gains High, low
Fan type Variable speed, outlet damper
Cooling plant Good, poor
68 HVAC&R RESEARCH
by 22.9 m), and oriented so that the longer sides faced the east and west directions. Windows comprised
20% of the wall area on all sides of the building.
The second building construction considered was a wood frame building of the same dimensions, orien-
tation, and window placement as the block building described previously. This building was designed to
represent lighter commercial construction. The exterior walls were modeled as exterior wood siding with
standard wood framing and fiberglass batt insulation. A cross-sectional view of the exterior wall is shown
in Figure 6. The window sizing, placement, and properties were identical to the block building. The roof
construction was also modified to reflect the lighter construction techniques. The floor used the same slab
on grade construction of the block building, but the tile flooring was replaced with carpet.
The third building construction was designed to represent a single story of a large multi-story curtain
wall construction building. The term curtain wall refers to the fact that the exterior walls are not structural
or load bearing so the walls are draped over the structural beams. Curtain wall construction is often seen
in high rise office buildings and typically features large glass areas. This building was modeled as an insu-
lated stone facade over reinforced concrete. Approximately 50% of the wall area was tinted low emmisivity
glass so there was high ambient coupling and large solar gains. Figure 7 shows a cross-sectional view of
the exterior wall construction. The model represented a single floor of a large building so the ceiling and
floor construction were identical. The zones above and below the floor were assumed to be at the same
thermal conditions as the test zone.
Finally, a zone which was not directly coupled to the environment was considered. The zone models a
single zone (900 ft
2
or 84 m
2
) within a multistory building with four interior gypsum board walls of metal
stud construction. The zone is coupled to identically-conditioned zones on all sides.
Internal gains are due to lighting, office equipment, and occupants. Two internal-gain profiles were
studied. The first internal gain schedule, termed low, uses the average office lighting and equipment val-
ues reported by Stein et al. (1986). The time distribution of the low internal gains is shown in Figure 8. The
lights and equipment were turned on from 7 A.M. to 7 P.M. The building was considered to be occupied
from 8 A.M. to 6 P.M. Gains from the occupants were determined using the average values for light office
work given in ASHRAE (1989a). An occupant density of one worker per 150 ft
2
(14 m
2
) was used. The
Figure 5. Block building exterior wall cross-sec-
tion
Figure 6. Frame building exterior wall cross-sec-
tion
VOLUME 2, NUMBER 1, JANUARY 1996 69
second internal gains schedule, termed high, used the same internal gains schedule, but the lighting loads
were doubled while the equipment loads were increased by 50%.
The cooling plants considered in this study cover the range of performance that would be seen in the
field (Braun 1990). In the context of this work, good cooling plants have better performance (higher COP)
as the part-load ratio drops; poor plants show the opposite effect. Good plant performance results from the
use of equipment having continuous capacity control (e.g., variable-speed drives on the chillers, pumps,
and fans). Poor plants are modeled after plants which consist entirely of fixed speed equipment.
Optimal Control Results
Figure 9 shows typical optimal and night setup cooling load profiles for a 70F (21C) day, clearness
index of 0.60, and a utility rate multiplier of 2.0. The cooling load profile under night setup control is given
by the dotted line. At night, the equipment was off and there was no load on the cooling plant. At occu-
pancy, the equipment was turned on and the resulting load profile peaked at approximately 3 P.M.
Figure 7. Multistory building exterior wall cross-section
Figure 8. Internal gains time distribution (low)
70 HVAC&R RESEARCH
Under optimal control, precooling occurred during the majority of the unoccupied period and stopped
just prior to occupancy to allow the zone conditions to reach the comfort zone at occupancy. Since the on-
peak period starts at 12 A.M., the large cooling rate at 11 A.M. served to shift load from the on-peak to the
off-peak period in an efficient manner. The area bounded by the two load profiles and the on-peak time
period is the cooling load which has been shifted by using the thermal mass of the structure. The load pro-
file during precooling could be broken into two distinct periods. The initial precooling occurred at a nearly
constant rate over the night period until two hours prior to occupancy. This was followed by a warmup time
with no precooling which allowed the zone conditions to return to the comfort region.
The zone PMV under the same simulation parameters is shown in Figure 10. During the unoccupied
night period, the zone was outside of the comfort limits. Figure 9 shows that precooling stopped two hours
before occupancy to allow time for warmup. Just prior to the initiation of the peak time-of-day (TOD) rate
period, the zone PMV was at the lower end of the comfort range. After the onset of the on-peak rates, the
PMV was at the upper limit of comfort which enabled the thermal mass to release the cooling energy stored
from the night period at the highest possible rate. These features were seen in the majority of the optimal
control simulations. Under optimal control, the entire range of occupant comfort was experienced during
the occupied period.
Two-Load Precooling Simplification
Results of the optimal cooling simulations were used to develop simplifications for the precooling
period. The optimal control results shown in Figure 9 led to simplifying the precooling to two periods of
cooling at a constant rate. The first period extends from the end of the previous occupied period until three
hours prior to occupancy. The second period covers the remaining three hours until the building is occu-
pied. The two-load precooling strategy is illustrated in Figure 11.
Examination of the optimal control strategies also led to a set of rules for controlling the space tempera-
ture during the occupied period. The rules, expressed in terms of occupant comfort, are as follows:
1. Prior to the onset of the on-peak utility rate, maintain the space at a value at the lower end of the comfort
Figure 9. Zone cooling under optimal control
(Block building, high gains, good plant, 70F (21C) day, k
t
= 0.60, R = 2.0)
VOLUME 2, NUMBER 1, JANUARY 1996 71
zone (PMV = 0.15) to save the discharge of the mass until the on-peak period.
2. When mechanical cooling is required during or after the on-peak period, maintain the space at the upper
end of the comfort limit (PMV = 0.5).
The overall effect of the simplifications is that the optimization problem has been reduced to adjust-
ing two constant zone sensible precooling rates ( and ) to minimize the daily cost. A flow chart for
the optimization procedure using the simplified optimal control is given in Figure 12. The two-load con-
Figure 10. Zone PMV under optimal control
(same system as used for results of Figure 9)
Figure 11. Two-load simplified control
Q

1
Q

2
72 HVAC&R RESEARCH
troller uses the two variables in conjunction with the occupied period comfort rules to generate the control
vector used to evaluate costs with Equation (6). The simplifications to the optimal control strategy still
require an optimization with two variables, but this is a significantly easier problem which was solved in
this study by a discrete search of the design space.
Figure 13 shows a plot of the performance of the simplified control compared to night setup and optimal
control over a range of average outdoor temperatures. In order to make relative comparisons between con-
trol strategies, all costs reported were normalized by the cost under night setup control using identical
parameters. The cost relative to night setup (F) is defined as:
(8)
where C is the daily cooling cost and C
ns
is the daily cooling cost under night setup control. Four different
control strategies were considered:
1. No precooling with comfort-based rules
2. Two load precooling with a fixed occupied T
z
of 75F (24C)
3. Two load precooling with comfort-based rules
4. Optimal control
Figure 13 shows that higher fractional savings can be achieved with building precooling on cooler days.
This is mainly due to the availability of relatively low cost ventilation cooling at night. It is important to
note that the absolute costs associated with both night setup and building precooling strategies increase as
ambient temperature increases. Figure 13 shows that the simplified method for estimating optimal control
works extremely well when the comfort-based rules are employed. However, simplified precooling with
constant day temperature set points (no rules) only results in about half of the savings compared to the opti-
mal control. A significant portion of the savings is due to the use of comfort-based rules as opposed to a
constant zone temperature daytime control. The comfort-based rules help delay the discharge of the ther-
mal mass until the onset of the on-peak utility rate period. This demonstrates that it is important to consider
both the charging and discharging of the thermal mass in the control strategy. When the comfort based
rules were used with no precooling, small savings were seen for all outdoor temperatures. The comfort-
based rules with no precooling shift a small amount of cooling load from the on-peak period by slightly
cooling the zone during the occupied off-peak utility rate period.
The simulated performance of night setup, optimal control, and simplified optimal control simulations
was compared using all possible combinations of the parameters shown in Table 1, with the exception that
the internal zone only used the high internal gains profile. This resulted in a total of 1008 different simula-
tions.
The cost relative to night setup using the simplified optimal control (F
simp
) was plotted as a function of
the optimal control cost ratio (F
opt
) for the systems described in Table 1 and is shown in Figure 14. The 45
line represents the best possible performance of the simplification where the cost under the simplifications
equals the cost under optimal control. Points close to this line are better approximations of optimal control
Figure 12. Optimization procedure using optimal control simplifications
F
C
C
ns
-------- =
VOLUME 2, NUMBER 1, JANUARY 1996 73
compared to more distant points. Figure 14 shows that the simplified optimal control provides an excellent
estimate of the optimal control in most cases.
A performance index was also defined to evaluate the overall performance of the simplifications to
optimal control. The performance index is defined as the ratio of cost savings for the simplified control to
that for the optimal control,
(9)
The overall performance index was found to be 0.97 for the 1008 daily simulations. The simplified
method performed poorly on only a relatively small number of days. Those days were predominantly cool,
with a high clearness index. A high clearness index corresponds to a large diurnal temperature swing. For
these cool days, the optimal control used night air for ventilation precooling and could adjust the cooling
rates at each time-step to account for the large variation in ambient air temperature. The simplified method
could not use ventilation cooling in the most efficient manner because only two constant precooling rates
can be specified.
Time-Load Precooling Simplification
A second simplification to the precooling period was evaluated which also uses two variables to approx-
imate the precooling strategy and the comfort-based rules during occupancy. This simplification is based
on applying a constant cooling rate for a given amount of time prior to building occupancy. A one hour
warmup period was included to allow the zone to return to the comfort region after precooling. The time-
load precooling strategy is illustrated in Figure 15. The flow chart of Figure 12 describes the optimization
with the two variables replaced by t
pre
, the duration of precooling prior to occupancy, and , the sensi-
Figure 13. Daily cooling cost
(Internal zone, Good plant, R = 2, k
t
= 0.75)

C
ns i ,
C
si mp i ,
( )
i 1 =
N

C
ns i ,
C
opt i ,
( )
i 1 =
N

------------------------------------------------- =
Q

pre
74 HVAC&R RESEARCH
ble cooling rate applied to the zone during precooling. Again, the optimal solution was determined by a
direct search of the control variables at discrete values.
The performance of this control strategy is similar to the two-load strategy as shown in Figure 16. The
performance index was found to be 0.95 over the range of simulations. The advantage of this precooling
strategy is that it is similar to the return from night setup control which attempts to estimate the minimum
time that the heating or cooling plant must turn on prior to occupancy in order to return the zone to the
Figure 14. Two-load precooling simplification scatter plot
Figure 15. Time-load simplified control
VOLUME 2, NUMBER 1, JANUARY 1996 75
comfort region. The return from night setup control has been extensively studied (Seem et al. 1989). It may
be possible to combine their techniques for return from night setup with the time-load simplification lead-
ing to an online controller for building precooling.
The time-load simplification was used to study the storage efficiency of the buildings considered in this
study. Some of the cooling performed at night will be lost to the ambient and will not directly offset cool-
ing during the day. The benefits of the precooling strategy (favorable utility rates, part load characteristics
etc.) must outweigh these losses. When discussing these effects, it is helpful to define a storage efficiency
for building precooling
storage
. The storage efficiency is the fraction of the cooling applied during the
night period that ultimately reduces the cooling load during the occupied period and is defined as:
(10)
where Q
shift
is the reduction in cooling load during the occupied period from precooling and Q
night
is the
total cooling applied during the precooling period.
To study the effect of the time of precooling on
storage
, precooling at a rate of one half of the design
load of the cooling plant was applied to the different building models using the time-load precooling sim-
plification shown in Figure 15. Figure 17 shows the storage efficiencies for the different building types as a
function of the time of precooling. The curtain wall building storage efficiency is higher than the block and
frame constructions because the curtain wall has adjacent zones above and below the test zone. As
expected, the storage efficiency for the heavier block wall building is higher than the frame building. The
magnitude of the change in the curtain wall building storage efficiency with precool time is about twice
that of the block and frame building. In general, the storage efficiency shows a stronger dependence on pre-
cooling time in buildings which are better coupled to the environment (block and frame). The storage effi-
ciency of the internal zone building model is 1.0 because the zone is completely surrounded by zones using
identical control strategies.
Figure 16. Time-load precooling simplification scatter plot

storage
Q
shift
Q
night
--------------- =
76 HVAC&R RESEARCH
Conclusions
This paper described two simplified approaches to the problem of determining the optimal control for
the use of building thermal mass in cooling. The simplifications are based on two variable approximations
to the precooling portion of the day, and a set of comfort-based rules when the building is occupied. The
first simplified approach approximates the precooling period by two periods of constant rate cooling. The
second approach uses a constant load applied over a specified time to describe the precooling period. The
simplifications were tested by examining 1008 different combinations of building types, weather condi-
tions, cooling plants, and utility rates. For each case, the optimal cost for cooling was calculated and com-
pared to the cooling cost using the optimal control simplifications. These simplifications were found to be
an excellent approximation to the optimal control, and yielded from 95% to 97% of the potential savings
relative to night setup control.
The simplifications are significant in two ways. First, they reduce the computation required for deter-
mining the daily optimal control for control strategy development. When using hourly set points, the opti-
mal control problem is reduced from 24 variables to two variables. The simplifications to the optimal
control problem also represent an important step in the development of on-line controllers. The computa-
tional requirements for an adaptive controller can be significantly reduced by using these simplifications.
For example, an on-line controller could use parameter estimation to learn a simplified model of the build-
ing thermal response. The two-variable optimization problem could then be solved and combined with the
discharge rules to completely describe the daily cooling strategy.
Acknowledgement
We wish to thank members of the project monitoring subcommittee of ASHRAE Technical Committee
TC 4.6, Building Operation Dynamics, for their guidance. The ASHRAE research support and additional
support provided by the Graduate Grant-In-Aid Program is greatly appreciated.
Figure 17. Building storage efficiency as a function of the time of precooling
(High gains, good plant, 80F (27C) day, k
t
= 0.60)
VOLUME 2, NUMBER 1, JANUARY 1996 77
NOMENCLATURE
A penalty function step constant
B penalty function rate constant
C daily cooling cost
C
ns
daily cooling cost under night setup control
C
opt
daily cooling cost under optimal control
C
sim
daily cooling cost under simplified optimal control
COP cooling plant coefficient of performance
F cost relative to night setup control
k time stage of simulation
k
t
daily average clearness index
N number of time stages in a day
fan power
Pen
pmv
penalty due to comfort constraint violation
PMV predicted mean vote comfort index
precooling rate during first precooling period
precooling rate during second precooling period
total load on the cooling coil
cooling plant capacity
zone latent cooling load
Q
night
total cooling during precooling period
constant precooling rate under time-load simplification
zone sensible cooling load
Q
shift
total cooling load shifted from the occupied period due to precooling
cooling load due to ventilation
R
k
on-peak utility rate multiplier defined as the ratio of on-peak to off-peak electricity cost
t
pre
precooling time for time-load precooling simplification
T
z
zone air temperature
T
s
supply air temperature to the zone
vector of control inputs to the building model
t simulation time step

storage
building mass storage efficiency
Optimal control simplification performance index
REFERENCES
Andresen, I., and M.J. Brandemuehl. 1992. Heat Storage in Building Thermal Mass: A Parametric Study. ASHRAE
Transactions 98(1): 910-918.
ASHRAE. 1989a. ASHRAE HandbookFundamentals. Atlanta: ASHRAE.
ASHRAE. 1989b. Ventilation for Acceptable Indoor Air Quality. ASHRAE Standard 62-1989. Atlanta: ASHRAE.
Braun, J.E. 1988. Methodologies for the design and control of central cooling plants, Ph.D. thesis, University of Wiscon-
sin-Madison.
Braun, J.E. 1990. Reducing Energy Costs and Peak Electrical Demand Through Optimal Control of Building Thermal
Mass. ASHRAE Transactions 96(2): 264-273.
Conniff, J.P. 1991. Strategies for Reducing Peak Air-Conditioning Loads by Using Heat Storage in the Building Struc-
ture. ASHRAE Transactions 97:(1) 704-709.
Duffie, J.A., and W.A. Beckman. 1991. Solar Engineering of Thermal Processes. New York: John Wiley and Sons.
Erbs, D.G. 1984. Models and applications for weather statistics related to building heating and cooling loads. Ph.D. the-
sis, University of Wisconsin-Madison.
Fanger, P.O. 1970. Thermal Comfort. Copenhagen: Danish Technical Press.
Golneshan, A.A., and M.A. Yaghoubi. 1990. Simulation of Ventilation Strategies of a Residential Building in Hot Arid
Regions of Iran. Energy and Buildings 14: 201-205.
Keeney, K.R., and J.E. Braun. 1995. Building Characteristics and Control Strategies for use of Building Thermal Mass
in Cooling. ASHRAE Technical Report 734-RP. Atlanta: ASHRAE.
Klein, S.A., et al. 1990. TRNSYS A Transient System Simulation Program, 13.1 edition. Solar Energy Laboratory, Uni-
versity of Wisconsin-Madison.
Knebel, D. 1983. Simplified Energy Analysis Using the Modified Bin Method. Atlanta: ASHRAE
Morris, F.B., J.E. Braun, and S.J. Treado. 1994. Experimental and Simulated Performance of Optimal Control of Build-
ing Thermal Storage. ASHRAE Transactions 100(1): 402-414.
Rabl, A., and L.K. Norford. 1991. Peak Load Reduction by Preconditioning Buildings at Night, International Journal of
Energy Research 15: 781-798.
P

fan
Q

1
Q

2
Q

coil
Q

cap
Q

l z ,
Q

pre
Q

s z ,
Q

vent
u
78 HVAC&R RESEARCH
Rao, S. 1984. Optimization. New Delhi, India: Wiley Eastern Limited.
Ruud, M.D., J.W. Mitchell, and S.A. Klein. 1990. Use of Building Thermal Mass to Offset Cooling Loads. ASHRAE
Transactions 96(2): 820-828.
Seem, J.E., P.R. Armstrong, and C.E. Hancock. 1989. Algorithms for Predicting Recovery Time From Night Setback.
ASHRAE Transactions 95(2): 439-446.
Snyder, M.E., and T.A. Newell. 1990. Cooling Cost Minimization Using Building Mass for Thermal Storage. ASHRAE
Transactions 96(2): 830-838.
Stein, B., J.S. Reynolds, and W.J. McGuinness. 1986. Mechanical and Electrical Equipment for Buildings. New York:
John Wiley and Sons.
Stephenson, D.G., and G.P. Mitalis. 1971. Calculation of Heat Conduction Transfer Functions for Multi-layer Slabs.
ASHRAE Transactions 77(2): 117-126.

You might also like