Está en la página 1de 24

VOLUME 13, NUMBER 4 HVAC&R RESEARCH JULY 2007

557
A Near-Optimal Control Strategy for
Cool Storage Systems with
Dynamic Electric Rates (RP-1252)
James E. Braun, PhD, PE
Fellow ASHRAE
Received October 31, 2006; accepted March 12, 2007
A near-optimal control method was developed for charging and discharging of cool storage sys-
tems when real-time pricing (RTP) electric rates are available The algorithm requires relatively
low-cost measurements (cooling load and storage state of charge), requires very little plant-
specific information, is computationally simple, and ensures that building cooling requirements
are always met (e.g., storage isnt prematurely depleted). The control method was evaluated for
ice storage systems using a simulation tool for different combinations of cooling plants, storage
sizes, buildings, locations, and RTP rates. The simplified method worked well in all cases and
gave annual costs within approximately 2% of the minimum possible costs associated with opti-
mal control.
INTRODUCTION
The primary objectives of the work described in this paper were to develop and evaluate a
simple control strategy for cool storage systems that works well with real-time pricing (RTP)
utility rate structures. RTP options are offered by many utilities to commercial and industrial
customers and incorporate time-varying rates that more closely approximate actual costs of pro-
viding service than conventional rate structures. The RTP rates vary hourly and are provided
electronically by utilities up to one day in advance of their application. Customers having cold
storage systems and energy management and control systems could incorporate controllers that
respond to RTP rates by adjusting charging and discharging rates of storage at different times of
day. However, no simple algorithms have been developed and demonstrated that provide
near-optimal control of thermal storage systems with RTP rates.
A number of simple control strategies have been developed for cold storage systems operat-
ing with conventional utility rates having on-peak and off-peak energy and demand charges
(Rawlings 1985; Tamblyn 1985; Spethmann 1989; Braun 1992; Drees and Braun 1996; Henze
2003a). All of these strategies assume that storage is fully charged prior to each occupied
period and incorporate chiller-priority and/or storage-priority discharge strategies. With
chiller-priority discharge, storage only operates to satisfy the difference between the load
requirement and the maximum chiller cooling capacity. This type of strategy minimizes the use
of storage and is appropriate in situations where the cost of operating the chillers is less than
the cost of recharging storage during the charging period. On the other hand, storage-priority
strategies maximize the use of storage during the on-peak occupied period. With a maxi-
mum-discharge storage- priority strategy, the chiller only operates to satisfy the difference
between the load requirement and the maximum storage discharge capacity. With a
James E. Braun is a professor of mechanical engineering at Purdue University, West Lafayette, IN.
2007, American Society of Heating, Refrigerating and Air-Conditioning Engineers, Inc. (www.ashrae.org). Published in HVAC&R Research, Vol. 13,
No. 4, July 2007. For personal use only. Additional reproduction, distribution, or transmission in either print or digital form is not permitted without
ASHRAEs prior written permission.
558 HVAC&R RESEARCH
demand-limiting or load-limiting storage-priority strategy, the discharge rate varies in order to
maintain a peak power or load at or below a specific target.
Relatively little work has been performed related to the development and evaluation of con-
trollers for cool storage systems in combination with RTP. Henze et al. (1997; Henze 2003b;
Henze and Krarti 1999) developed a predictive optimal controller that determines optimal trajec-
tories for charging and discharging storage based upon minimizing an integral cost function with
RTP rates. Implementation of the method requires predictive models for the cooling plant and
cool storage, forecasts for cooling loads and ambient temperatures, and knowledge or forecasts
of RTP rates some hours in advance. Henze et al. (1997) showed that optimal control results in
very significant savings for systems with RTP rates as compared with conventional strategies
that are typically employed for systems with conventional time-of-use energy and demand
charges. However, the requirement of developing/configuring models for each application is
probably cost-prohibitive, and there is a need for a simple strategy that performs well in relation
to optimal control.
This paper develops and validates a simple method for controlling the charging and discharg-
ing of thermal storage systems having RTP utility rates. The development starts with the method
of Drees and Braun (1996), a modified version of which is described in chapter 41 of the 2003
ASHRAE HandbookHVAC Applications. The original strategy was developed for systems
with conventional time-of-use rates that employ discrete on-peak and off-peak energy and
demand rates. The strategy switches between chiller-priority, maximum-discharge storage-pri-
ority, and load-limiting storage-priority strategies based upon economics and availability of
storage and uses forecasting to ensure against premature depletion of storage. The primary con-
tributions of the current paper are (1) the recognition that the method of Drees and Braun (1996)
can be applied to systems having continuously varying RTP rates by using effective on-peak and
off-peak periods, (2) development of a method for determining the effective on-peak and
off-peak periods, and (3) an extensive simulation evaluation of the method for different combi-
nations of cooling plant configurations and sizes, thermal storage sizes, buildings, locations, and
RTP rates. A benchmark optimization tool was developed to determine the true optimal con-
trol for the simulated systems, and the performance of the simplified algorithm is measured rela-
tive to the benchmark.
SYSTEM MODELING
Figure 1 shows a schematic of the system configuration considered in this paper. This system
is representative of configurations employed for ice storage systems having internal-melt,
ice-on-pipe tanks with water-cooled chillers. Air-cooled chillers are also considered in this study
because they are common for medium and light commercial applications. Internal-melt,
ice-on-pipe storage tanks are the most common type of thermal storage system. A mixture of
water and glycol is the working fluid that circulates between the building and the cooling plant.
At any given time, the building load can be met by running the chiller or by depleting the storage
or by a combination of the two. The chiller loading is dictated by the setpoint for the supply tem-
perature from the chiller (T
chws
). The ice storage tank provides any additional cooling necessary
to maintain the required supply temperature to the building cooling coils (T
coil
). During charg-
ing mode, the chiller supply and load supply temperature setpoints are set to low values below
the freezing point of water to ensure that the chiller operates at full capacity, all of the flow goes
through storage, and ice is produced. Charging generally occurs during the unoccupied period
when the load requirements are either nonexistent or are small and most or all of the plant
water/glycol supply flow is short-circuited to the return line.
Sun et al. (2006) provide a detailed description of the simulation tool used in this study,
whereas this section provides a brief overview of the modeling approach and plant characteristics.
VOLUME 13, NUMBER 4, JULY 2007 559
Ice Storage Tanks
The storage tanks are considered to be internal-melt, ice-on-pipe configurations and are mod-
eled using the semi-empirical model developed by West and Braun (1998). The storage model
neglects any heat gains through the storage shell and assumes that the state of storage can be
represented with a single variable that defines the fraction of the maximum available storage
capacity. For any time interval k,
(1)
where Cap
s
is the maximum change in internal energy of the storage tank that can occur during
a discharge cycle, u is the rate of energy addition to storage over the stage (positive for charging
and negative for discharging), and t is the simulation time step. The state of charge defined in
this manner must be between zero and one. In this study, zero state of charge corresponds to a
tank of water at a uniform temperature of 32F (0C) and a complete charge is associated with a
tank that is frozen solid at 32F (0C).
The rates of charging and discharging of storage are limited by the storage heat exchanger
area, secondary fluid flow rate and inlet temperature, and the thickness of ice at any time. The
maximum charging and discharging rates are determined using a heat transfer effectiveness
model described by West and Braun (1998) for internal melt, ice-on-pipe storage tanks. At any
stage, the maximum charging and discharging (minimum charging) rates are determined as:
(2)
(3)
where
c,k,max
and
d,k,max
are heat transfer effectiveness for charging and discharging at the cur-
rent state of storage if all of the chilled-water/glycol flow passes through the tank (no bypass)
with a flow rate of , c
f
is the specific heat of the water/glycol mixture flowing through
chiller and storage, T
chws
is the chiller supply temperature that is inlet to the tank, and T
s
is the
temperature at which the storage medium melts or freezes (32F [0C] for this study). Storage
heat transfer limitations and the use of these effectiveness models only come into play when the
Figure 1. Schematic of ice storage system.
x
k
x
k 1
u
k
t
Cap
s
------------ , + =
u
k, max

c, k,max
m

chw
c
f
T
s
T
chws
( ) =
u
k,min

d, k, max
m

chw
c
f
T
s
T
chws
( ) =
m

chw
560 HVAC&R RESEARCH
all of the secondary flow is through storage and there is no bypass. In other situations, the stor-
age control (three-way valve in Figure 1) bypasses some flow to maintain a load supply temper-
ature, and the storage heat transfer rate is less than the maximum possible.
The method of West and Braun (1998) uses effectiveness data determined from experiments
for complete charging and discharging cycles where a tank goes from fully discharged to fully
charged and is then returned to a fully discharged state. Charging and discharging effectiveness
are then correlated as functions of the state of charge. Figure 2 shows example heat transfer
effectiveness results for full charging and discharging cycles from Drees and Braun (1995) that
were employed in the current study. The effectiveness is relatively constant over the first
60%70% of the charging cycle. In this range, ice formations building on adjacent tubes do not
intersect and the heat transfer effectiveness drops slowly as the thickness of ice grows. During
the last quarter of the charging cycle, the effectiveness drops quickly due to the rapid loss in
surface area at the ice/water interface caused by overlapping ice formations. During the dis-
charge cycle, ice near the tube melts and forms a water layer. The effectiveness falls with
increasing water layer thickness because heat transfer is dominated by conduction for these
thin layers and the water has about one-third the thermal conductivity of ice. During the last
quarter of the discharging cycle, the effectiveness drops more quickly due to the rapid loss in
heat transfer area. In the current study, the minimum and maximum storage states were set as
0.1 and 0.9, respectively, in order to ensure adequate heat transfer rates during charging and
discharging,. The lower limit corresponds to a tank with 7% ice by mass at a uniform tempera-
ture of 32F (0C), while the upper limit is associated with a tank condition where the ice for-
mations intersect slightly.
Although the effectiveness results of Figure 2 were developed for complete charging and dis-
charging cycles, the method of West and Braun (1998) corrects the data to predict effectiveness
for any scenario that involves combinations of partial charging and discharging of storage. In the
current study, only a single charging period was considered and the state of charge was returned
Figure 2. Full charging and full discharging effectiveness curves for internal melt,
ice-on-pipe storage at 50% of design flow rate.
VOLUME 13, NUMBER 4, JULY 2007 561
to the upper charge limit of 0.9 at the end of the period. Furthermore, no discharging of storage
could occur during the charging period. These assumptions eliminated the possibility of multiple
trapped layers of ice and/or water occurring due to arbitrary sequences of partial charging and
discharging cycles. However, partial discharging of storage can occur so that a single layer of
water could be trapped between ice formations during the charging period.
Chiller Plants
Drees and Braun (1996) developed an empirical model for an ice-making chiller plant
employing a centrifugal chiller that was utilized in the current study. The model includes a cor-
relation for plant cooling capacity as a function of chiller supply temperature and ambient
wet-bulb temperature and a correlation for plant power consumption as a function of chiller
cooling load, chiller supply temperature, and ambient wet-bulb temperature. The plant power
consumption includes power for the chiller, condenser water pump, and cooling tower fan.
Chilled-water/glycol pumping power is treated separately. The correlations were developed
from results of simulations that incorporated individual chiller, cooling tower, and condenser
water pump component models. The simulation model incorporated a chiller model developed
from manufacturers data for an ice-making centrifugal chiller. The cooling tower was modeled
using effectiveness relations that account for heat and mass transfer with a single-speed fan. The
condenser pump had constant water flow and power consumption when operating.
Figure 3 shows plant cooling capacity relative to rated capacity as a function of chiller supply
temperature and ambient wet-bulb temperature. The capacity is very sensitive to the chiller sup-
ply temperature and varies by a factor of 2 between conditions associated with ice-making and
ice-melting conditions. The sensitivity to ambient wet bulb is less, but still significant.
Figure 4 shows overall cooling plant efficiency as a function of load relative to rated capacity
and difference between ambient wet-bulb and chiller supply temperatures. The efficiency
degrades significantly as the load decreases due to two effects. First, chiller efficiency is typi-
cally best at full load for chillers that rely on inlet vane control for reducing cooling capacity.
Even more important, the auxiliary power associated with the tower fan and condenser pump is
Figure 3. Relative cooling capacity for water-cooled centrifugal chiller plant.
562 HVAC&R RESEARCH
constant with changes in load and is a much larger percentage of the total power consumption at
low loads where chiller power is less.
In a sense, the results of Figure 4 represent a lower limit on the part-load performance of a
chiller plant because the plant only incorporates a single stage of control for the chiller, con-
denser pump, and cooling tower fan. An alternative part-load performance characteristic was
developed by considering a plant that has infinite stages of operation for multiple chillers, con-
denser water pumps, and cooling tower cells and fans. It was assumed that the overall plant
power varies linearly with loading and the overall plant efficiency is independent of load and
equal to the efficiency when the plant is operating at full capacity. For this case, Figure 5 shows
the dependence of overall plant efficiency on ambient wet-bulb and chiller supply temperatures,
independent of the load.
Air-cooled chillers are commonly employed for ice-making applications involving
medium-size and light commercial buildings because of lower initial and maintenance costs
compared to water-cooled systems. For medium-size applications, air-cooled screw chillers
are commonly used, whereas air-cooled scroll chillers are typical for light commercial build-
ings. Data from a manufacturer were acquired for air-cooled screw and scroll chillers
employed in ice storage systems. However, the data for both chiller types were quite similar
so only a screw chiller model was considered in this study. The data were used to develop
empirical models for cooling capacity and power consumption of the chiller and condenser
fans operating at full capacity as a function of ambient temperature and chiller supply temper-
ature (see Sun et al. [2006]).
Figure 6 shows plant cooling capacity relative to rated capacity as a function of chiller supply
temperature and ambient wet-bulb temperature for the air-cooled screw chiller. The sensitivity
of capacity to supply temperature is similar to that for a centrifugal chiller and varies by a factor
of 2 between conditions associated with ice-making and ice-melting conditions.
Figure 4. Plant efficiency for single-stage centrifugal chiller plant (0.66 kW/ton at rating
condition).
VOLUME 13, NUMBER 4, JULY 2007 563
Figure 5. Plant efficiency for infinite-stage centrifugal chiller plant (0.66 kW/ton at rating
condition, no part-load dependence).
Figure 6. Relative cooling capacity for air-cooled screw chiller.
564 HVAC&R RESEARCH
Figure 7 shows overall efficiency as a function of chiller supply temperature and ambient
wet-bulb temperature for the screw chiller. According to manufacturers data, the chiller effi-
ciency (including condenser fans) is about 1.16 kW/ton at the rating condition. This is signifi-
cantly higher than the rated value for the centrifugal chiller. Furthermore, the effect of ambient
and chilled-water conditions on plant efficiency is greater than for the centrifugal chiller over
the operating range of interest.
Chilled-Water/Glycol Distribution, Loads, and Setpoints
The flow rate and pump power associated with chilled-water/glycol distribution are assumed
to be constants whenever there is a cooling load. Water flows and power are scaled with chiller
cooling capacity according to user-specified rating values for volumetric flow rate per unit rated
cooling capacity (gpm/ton or L/skW) and pump power per unit volumetric flow rate (W/gpm or
Ws/L).
The rate of charging (or discharging) of storage is a control variable that was determined by
the benchmark optimization or a simplified control strategy in this study. In general, the chiller
load at any time interval is the sum of the building cooling load and the storage charging rate
(negative for discharging) or
(4)
where is the chiller cooling load and is the total plant cooling load (building load).
Storage is typically charged during unoccupied times when the chiller load is small or nonex-
istent and is controlled by setting a low chiller supply set temperature. Discharge (negative charg-
ing) occurs during occupied periods when the electric rates are high and is controlled by setting a
relatively high chiller supply temperature. For any time interval, the charging/discharging rate
must be within bounds dictated by the heat transfer limitations of storage and chiller cooling
Figure 7. Overall efficiency for air-cooled screw chiller (1.16 kW/ton at rating condition;
includes condenser fans, no part-load dependence).
Q

ch
Q

load
u
k
, + =
Q

ch
Q

load
VOLUME 13, NUMBER 4, JULY 2007 565
capacity constraints (between 0 and maximum cooling capacity). These constraints are included
within the solutions for both the benchmark optimization and simplified control strategy.
The chiller supply temperature setpoint (T
chws
) depends on the operational mode (charging or
discharging), the temperature setpoint for supply to the load (T
coil
), and the storage charging or
discharging rate. During storage discharge, the flow through storage is modulated using a
three-way valve (see Figure 1) to maintain a specified load supply temperature (T
coil
). In this
case, the chiller supply temperature setpoint (T
chws
) is adjusted to regulate the discharge rate
according to
(5)
Except during charging mode, the load supply temperature setpoint (T
coil
) was assumed to be
constant at 40F (4.4C) for all simulations in this study.
During charging of storage, the load supply setpoint (T
coil
) is set below the chiller supply tem-
perature (T
chws
) and the storage temperature (T
s
) so that the control valve diverts all of the flow
through storage. In this case, the chiller supply temperature setpoint (T
chws
) is adjusted to
achieve a specified charging rate according to
(6)
For safe operation of the chiller, the setpoint is constrained between minimum and maximum
setpoints of 20F and 65F.
Building Loads and Weather Conditions
Hourly cooling plant load requirements for different buildings and locations were determined
using DOE-2 simulations and were incorporated into the simulation tool along with the coinci-
dent TMY2 weather data (NREL 1995) for dry-bulb and wet-bulb temperature. Four different
building types (office, retail store, school, and warehouse) were considered in four locations
(Chicago, Houston, New York, and San Francisco). Temperature setback was applied during
unoccupied periods for all the buildings in the simulation. Table 1 shows a brief description of
the simulated buildings. Figure 8 shows example cooling load variations for the different build-
ings on weekdays in Chicago. There are significant differences in the load shapes for these
Table 1. General Description of Simulated Buildings
Office Retail Store School Warehouse
Floor area, ft
2
100,000 130,000 80,000 50,000
Number of floors 4 3 2 1
Period for building
occupancy and
ventilation
Mon.Fri.
6 a.m.10 p.m.
Mon.Sat.
6 a.m.10 p.m.
Sun.
10 a.m.8 p.m.
Mon.Sun.
6 a.m.10 p.m.
Mon.Fri.
6 a.m.6 p.m.
Sat.
8 a.m.4 p.m.
Period for normal
thermostat settings
Mon.Fri.
8 a.m.5 p.m.
Mon.Sat.
9 a.m.9 p.m.
Sun.
10 a.m.7 p.m.
Mon.Fri.
8 a.m. p.m.
Mon.Fri.
8 a.m.5 p.m.
Sat.
8 a.m.4 p.m.
T
chws
T
coil
u
k
m

chw
c
f
----------------- . =
T
chws
T
s
u
k

c,max
m

chw
c
f
--------------------------------- . =
566 HVAC&R RESEARCH
buildings. The office building has relatively high loads during occupied periods and has practi-
cally no load during other times. The retail store has a more pointed load shape that peaks in the
late afternoon. The school has a relatively flat load profile during the school day and additional
loads that occur during periods of activities on evenings and weekends. The warehouse has a rel-
atively flat load and short occupancy. A detailed description of the simulated buildings is given
by Sun et al. (2006) along with several examples of the load variations.
RTP Rates
A model was developed for characterizing the time dependence of typical RTP rates that is
described in detail by Sun et al. (2006). The model produces a time-varying price for the costs of
electricity that depends on time of day and maximum temperature for the day (T
max
). Different
models were developed from RTP rates for two different utilities located in two different regions
within the US. For these two utilities, four different models were developed for weekdays and
weekends in both summer and winter. The rate models for summer periods were implemented in
the simulation tool and simulations were limited to periods between April and October.
Figures 9 and 10 show outputs from these models on moderate, hot, and very hot days. Utility 1
generally has higher weekday rates with peaks that occur earlier in the afternoon compared to
Utility 2. However, Utility 2 has higher rates on hot weekend days. The peak rates increase dra-
matically with peak temperature for both utilities.
Model Implementation and Equipment Sizing
The chiller plant and storage models were implemented in a computer program that reads
hourly load and ambient data and applies storage control strategies. The cooling plant models
provide predictions of total cooling plant power requirements for the chiller, pumps, and fans
(cooling tower or air-cooled condenser) as a function of loading and ambient conditions
(wet-bulb or dry-bulb temperature) at any hour. The hourly chiller load is dictated by the building
Figure 8. Average load profiles for Chicago on weekdays in July.
VOLUME 13, NUMBER 4, JULY 2007 567
load requirements and the storage discharging or charging rate. The building loads were deter-
mined for different building types and locations using DOE-2 simulations.
The storage discharging and charging rates depend on the control strategy. For the benchmark
optimization, both charging and discharging rates are determined by an optimizer (see next sec-
tion) that includes constraints for storage states and heat transfer rates and chiller cooling capac-
ity and supply temperature. The simplified control approach involves the use of chiller-priority
and storage-priority control strategies. For chiller-priority control during discharge, the chiller
supply setpoint temperature is set equal to the load supply temperature (40F [4.4C]). If the
chiller cooling capacity were not sufficient to maintain the setpoint, then the three-way valve for
the system of Figure 1 would modulate the discharge rate from storage. In this case, Equations 4
and 5 are solved along with the correlation for chiller capacity in order to determine the chiller
supply temperature, chiller cooling capacity/rate, and storage discharge rate that satisfy the load
while minimizing the use of storage. Otherwise, the storage discharge rate is zero and the chiller
cooling rate is equal to the building load (Equation 4).
For storage-priority control, the chiller supply setpoint should be adjusted to minimize
the chiller load while satisfying the load and a constraint on discharge rate from storage.
For the model, this involves the solution of Equations 4 and 5 along with chiller and load
Figure 9. RTP rates for weekdays.
Figure 10. RTP rates for weekend days.
568 HVAC&R RESEARCH
energy balances to determine discharge rate, chiller supply temperature, chilled-water/gly-
col return temperature, and chiller load. If the discharge rate exceeds the maximum dis-
charge rate dictated by Equation 3, then Equations 3, 4, and 5 are solved for discharge rate,
chiller supply temperature, and chiller load.
The simplified control strategy involves charging at the maximum cooling capacity of the
chiller where all of the secondary fluid flows through the tank and the load supply temperature
floats to a value dictated by the limitations on heat transfer for storage and chiller capacity. The
model first determines chiller and load supply temperatures iteratively by solving equations for
maximum chiller cooling capacity (Figures 3 and 6), maximum storage charging rate (Equation 2),
and energy balances on the chiller and load. If the chiller supply temperature is below the lower
limit, then the equations are solved without the capacity equation and with the chiller supply tem-
perature set to the lower limit.
The loads determined with DOE-2 simulations were scaled to give a peak load of 400 tons
(1407 kW) for each building and location. There are a variety of different combinations of
chiller and storage size that will meet this peak building load requirement and a range was con-
sidered in parametric studies. In these studies, the storage size is referenced to the storage size
that results in the smallest possible chiller capacity. The program determines the minimum
chiller capacity and associated storage requirement by analyzing a design day that is chosen as
the day with the highest integrated load. An optimization is performed on the design day to
determine the storage size that gives the minimum possible rated chiller cooling capacity. For
discharging, a chiller-priority strategy is employed for this design-day sizing, and the full extent
of the storage capacity between the minimum and maximum storage states (0.1 and 0.9) is uti-
lized. For each iteration of the storage size optimization, the chiller capacity is determined that
meets both the peak cooling load and the charging rate necessary to fully recharge the storage
within a specified charging period. For parametric studies, the storage size is specified as a per-
centage (50% to 200%) of the size that gives minimum chiller capacity. Also, oversizing of the
chiller between 10% and 30% is considered.
BENCHMARK OPTIMIZATION
The benchmark optimization problem is defined by minimizing utility costs for the cooling
plant over the cooling season. For a given time horizon, the optimization problem can be stated
mathematically as follows.
Minimize
(7)
subject to satisfying the state equation (Equation 1) and load requirements (Equation 4) along
with the following control variable, state variable, and equipment constraints.
(8)
(9)
(10)
(11)
(12)
J R
k
P
k
t
k 1 =
N

=
u
min, k
u
k
u
max,k

x
min
x
k
x
max

T
chws, min
T
chws,k
T
chws, max

0 Q

ch, k
Q

ch, max,k

x
N
x
0
=
VOLUME 13, NUMBER 4, JULY 2007 569
where J is the utility cost for the optimization period, t is stage time interval (e.g., one hour), N is
the number of time stages in the optimization period, and for each state k, P is the plant electrical
power (kW), R is the cost per unit of electrical energy ($/kWh), u is the rate of energy removal
from storage over the stage (positive for charging and negative for discharging), x is the fractional
state of storage at the end of the stage, is the chiller cooling load, T
chws
is the chiller supply
setpoint temperature, and the subscripts min and max refer to lower and upper limits.
The utility rate at any stage is dictated by the RTP utility rate model. The plant power con-
sumption depends on the chiller cooling load, chiller supply setpoint temperature, and ambient
conditions (dry bulb or wet bulb). Both the chiller load and chiller supply setpoint temperature
are dictated by the building load and rate of storage charging or discharging. The state of charge
is updated for each interval using the state equation.
The limits for the constraint on the rate of charging/discharging of storage (Equation 8) are
due to heat transfer limitations dictated by the effectiveness models. Limits on the fraction state
of charge (Equation 9) were set to 0.1 and 0.9 to ensure adequate heat transfer at the extremes of
tank operation. The constraint of Equation 10 is used to ensure safe operation of the chiller, and
the limits were set to 20F and 65F. The chiller capacity models are used to limit chiller load
according to the constraint of Equation 11. Finally, the constraint of Equation 12 is enforced to
remove the effects of any assumption regarding initial state of the storage.
The benchmark optimization was performed using a series of 24-hour optimizations that
began each day at the time associated with the normal occupied period. Furthermore, storage
was assumed to be fully charged at the start of each optimization period (i.e., x
0
= 0.9) and per-
fect knowledge of load, weather, and RTP rates was used to determine the benchmark optimiza-
tion results. Each daily optimization problem was solved in a discrete manner using an iterative
form of dynamic programming (Sun et al. 2006). This method handles constraints on both state
and control variables in a straightforward manner and also guarantees a global minimum.
SIMPLIFIED RTP CONTROL STRATEGY
It would be possible to employ a numerical solution to the optimization problem posed in
Equation 7 in an online application. However, this approach would require the development or
configuration of models for each application, which would probably be cost-prohibitive. A plant
model could either be configured using information collected from the site or trained using mea-
sured data. The first approach would require time-consuming collection of detailed information,
whereas the second approach would require expensive power consumption measurements.
There were several goals associated with the development of a simple controller for cold stor-
age systems with RTP rates. First, the controller should have low hardware costs. This implies
that there should be minimal additional sensor requirements (e.g., no power measurements) for
implementation of the controller. Second, the cost for configuration of the controller should be
low. Therefore, the method should not utilize detailed plant models that might require signifi-
cant input requirements or substantial training. Third, the method should be robust and ensure
that building cooling requirements are always met (e.g., storage isnt prematurely depleted).
Fourth, the controller should have good performance in terms of operating costs as compared to
the benchmark optimal costs.
With these goals in mind, results from the benchmark optimization/simulation tool were used
to develop/evaluate simple heuristics that could be employed within a simple control strategy.
For charging of storage, it was found that a very simple, near-optimal strategy is to fully
recharge storage with the chiller operating at maximum capacity during a period defined by
when the RTP rates are lowest and the building is unoccupied. For discharging of storage, it was
found that the best strategy is to use a storage priority control that maximizes the discharge rate
of storage during a period defined by when RTP rates are highest, the building is occupied, and
Q

ch
570 HVAC&R RESEARCH
it is economical to utilize storage. For all other times, it is best to use chiller-priority control that
minimizes the discharge rate of storage.
These charging and discharging heuristics can be cast within the framework of the control
strategy of Drees and Braun (1996) that is described in chapter 41 of the 2003 ASHRAE Hand-
bookHVAC Applications (ASHRAE 2003). This strategy switches between chiller and storage
priority strategies, based upon economics and availability of storage, and ensures against prema-
ture depletion of storage through load forecasting. The original strategy was developed for cool
storage systems having conventional utility rates with on-peak and off-peak energy and demand
charges. For RTP rates, the same strategy is employed with no demand charges and with effec-
tive on-peak and off-peak periods for energy charges.
Chiller and storage load profiles for the overall control strategy are illustrated in Figure 11
along with the effective on-peak period. Maximum discharge storage-priority control is used
throughout the effective on-peak occupied period and chiller-priority control is employed at
other times. The effective on-peak period satisfies the following three criteria:
1. The RTP rates during the effective on-peak period must be greater than a threshold for eco-
nomic discharge of storage defined as
(13)
where RTP
c
is an RTP rate for the charging period, COP
c
is the cooling plant coefficient of
performance (load/power) associated with charging storage, and COP
d
is the plant COP asso-
ciated with discharging of storage under a chiller-priority strategy.
2. The effective on-peak period is the maximum possible time period where the chillers can
remain off.
3. The effective on-peak period results in the maximum possible average RTP rates subject to
the constraints specified by the first and second criteria.
Figure 11. Illustration of control strategy for cold storage control with RTP rates.
RTP
th
RTP
c

COP
d
COP
c
-------------- , =
VOLUME 13, NUMBER 4, JULY 2007 571
The first criterion, expressed in Equation 13, results from equating the cost per unit of cooling
required for storage-priority control to the cost per unit of cooling if the chiller is the primary
choice for meeting the load (chiller-priority control). The second and third criteria minimize the
use of storage until it is most cost-effective and then minimize the use of the chillers when their
costs are highest. The strategy also loads chillers at the maximum possible rate when they are
needed, which tends to give maximum chiller efficiency.
This simple strategy works well because (1) chiller plants having fixed-speed equipment typi-
cally operate most efficiently at full load, (2) there is very little penalty associated with recharg-
ing internal melt, ice-on-pipe storage to full charge, (3) storage is only depleted when it is
economical to do so, and (4) RTP rates are typically lowest during unoccupied times.
In order to implement the control strategy, it is necessary to have measurements of building cool-
ing load requirements, ambient conditions (dry bulb for air-cooled plants or wet bulb for
water-cooled plants), and the state-of-charge of storage. In addition, it is necessary to have
knowledge of RTP rates, forecasts of hourly cooling loads for the next occupied period, and
methods for estimating COP
c
and COP
d
. The following subsections address specific implemen-
tation issues for the simplified control strategy.
Charging
The charging period begins at the earliest possible time after the end of building occupancy
when the effective off-peak period is enabled (i.e., after the effective on-peak period has ended)
and ends when either storage is fully charged or the building occupancy period begins. The
charging strategy involves operating the chillers at maximum cooling capacity and would be
implemented by setting the chiller and load supply temperature setpoints to the lowest accept-
able value (e.g., 20F [6.6C]). With this approach, all of the secondary fluid flows through the
tank and the load supply temperature floats to a value dictated by the limitations on heat transfer
for storage and chiller capacity.
Chiller-Priority Control
Chiller-priority discharge is enabled during the occupied, effective off-peak period. During
this time, the chiller supply setpoint temperature is set equal to the load supply temperature set-
point (40F [4.4C]). If the chiller cooling capacity were not sufficient to maintain the setpoint,
then the three-way valve for the system of Figure 1 would modulate the discharge rate from stor-
age to maintain the load supply temperature setpoint.
Maximum-Discharge Storage-Priority Control
Maximum-discharge storage-priority control occurs during the effective on-peak period and
is implemented by turning the chillers off. The end of the effective on-peak period is continually
updated to avoid premature depletion of storage.
Determining the Start Time for the Effective On-Peak Period
At each decision interval (e.g., t
k
= 10 minutes) during the occupied period and before the
effective on-peak period has been initiated, the following steps are applied to determine the start
time for the on-peak period:
1. Evaluate RTP
th
using Equation 13. RTP
c
should be determined using utility rates for the time
when the most recent charging period ended. Typically, this is an RTP base rate that doesnt
vary from day to day. The ratio COP
d
/COP
c
should be evaluated using plant performance
information and typical charging and discharging conditions. The charging COP should be
572 HVAC&R RESEARCH
evaluated at full-load conditions with a typical chiller supply setpoint (e.g., 20F [6.6C])
for ice storage systems) and an ambient temperature determined when the most recent charg-
ing period ended. The discharging COP should also be evaluated under full-load conditions
(i.e., chiller-priority control) with a typical chiller supply setpoint (e.g., 40F [4.4C]) and the
current ambient condition.
2. Obtain known or determine forecasted electricity rates for each interval of the rest of the
occupied period, R
k
for k = k
I
to k
N
.
3. Determine building cooling load forecasts for each interval of the rest of the occupied period,
for k
I
to k
N
. Chapter 41 of the 2003 ASHRAE HandbookHVAC Applications pre-
sents an algorithm for forecasting cooling loads that could be applied.
4. Scan the electricity rates and find the first and second intervals where R
k
= RTP
th
. Assign
these as the minimum on-peak start interval, k
on,min
, and maximum on-peak end interval,
k
off,max
.
5. Scan the electricity rates and determine the time interval having the highest electricity rate
for the occupied period, m. Set the initial start and end intervals for the on-peak period as
k
on
= m and k
off
= m + 1.
6. Use the state-of-charge at the beginning of the current interval and forecasted loads and esti-
mates of chiller cooling capacity to estimate the state-of-charge at the start of the effective
on-peak period assuming chiller-priority control according to
7. Use forecasted loads to estimate the state-of-charge at the end of the current effective
on-peak period assuming storage-priority control according to
8. If then exit the algorithm. Otherwise, go to step 9.
9. If ( and and ) then set and go
to step 6. Otherwise go to step 10.
10. If ( and ) then set and go to step 6. Otherwise exit
the algorithm.
Figures 12 and 13 show example values for COP
d
/COP
c
determined from the models of the
water-cooled and air-cooled chiller plants as a function of the ambient temperatures (wet bulb
for water-cooled plant and dry bulb for air-cooled plant) occurring for charging (T
amb,c
or T
wb,c
)
and discharging (T
amb,d
or T
wb,d
). The chiller setpoint temperatures for charging and discharging
were assumed to be 20F (6.6C) and 40F (4.4C). The COP ratio is in the range of 0.9 to 1.3
and has a relatively strong dependence on ambient conditions, particularly for the air-cooled
chiller. For the current study, correlations were developed from these results and implemented
within the simplified control strategy.
Determining the End Time for the Effective On-Peak Period
Once the effective on-peak period has been initiated, then maximum-discharge storage-priority
control is utilized until the end of this period. An estimate of the end time is determined during the
Q

load, k
x
k
on
x
k
I
max Q

load,k
Q

ch, max, k
, ( ) Q

ch, max,k
( ) t
k

Cap
s
------------------------------------------------------------------------------------------------------- .
k k
I
=
k
on
1

=
x
k
off
x
k
on
Q

load, k
t
k

Cap
s
------------------------- .
k k
on
=
k
off
1

=
x
k
off
x
min
<
k
off
1 + k
off, max
< R
k
off
1 +
R
k
on
1
> Q

load,k
off
1 +
0 > k
off
k
off
1 + =
k
on
1 k
on, min
k
on
1 k
I
k
on
k
on
1 =
VOLUME 13, NUMBER 4, JULY 2007 573
process of determining the start time. However, it is necessary to continually update this estimate
throughout the period in order to ensure against premature depletion of storage and to utilize more
up-to-date forecasts. At each decision interval (e.g., t
k
= 10 min) during the effective on-peak
period, the following steps are applied to update the end time for the on-peak period:
1. Update RTP
th
using the current ambient temperature for the discharge condition.
2. Obtain known or determine forecasted electricity rates for each interval of the rest of the
occupied period, R
k
for k = k
I
to k
N
.
Figure 12. COP ratios for plant with water-cooled centrifugal chiller.
Figure 13. COP ratios for plant with air-cooled screw chiller.
574 HVAC&R RESEARCH
3. Determine building cooling load forecasts for each interval of the rest of the occupied period,
for k
I
to k
N
. To ensure sufficient cooling capacity, a worst-case forecast of cooling
requirements should be estimated as the sum of the expected value and two or three times the
standard deviation of the errors of previous forecasts.
4. Scan the electricity rates for the rest of the occupied period and find the next interval where
R
k
= RTP
th
. Assign the maximum on-peak end interval as k
off,max
.
5. Set k
off
= k
I
+ 1.
6. Use forecasted loads to estimate the state-of-charge at the end of the current effective
on-peak period assuming storage-priority control according to
(14)
7. If ( and k
off
+ 1 < k
off,max
) then set k
off
= k
off
+ 1 and go to step 6. Otherwise exit the
algorithm.
Controller Behavior and Performance
In order to compare controller performance, both the optimal and simplified control strategies
were implemented within the simulation environment with the assumption of perfect forecasts
of cooling loads and RTP rates. As a first step in illustrating the behavior of the optimal and
near-optimal control, a single day was simulated with a simple load profile and the following
specific parameters and conditions: (1) single-stage, water-cooled centrifugal chiller plant,
(2) rectified sinusoidal variation of cooling load with a mid-afternoon peak of 400 tons (1407
kW), (3) sinusoidal variations in wet-bulb and dry-bulb temperatures with minimum values of
65F (18.3C) and 85F (29.4C) at 2:30 a.m. and maximum values of 75F (23.9C) and 95F
(35C) at 2:30 p.m., (4) storage size set to 200% of the storage size that leads to minimum chiller
size, (5) 10% chiller oversizing, and (6) RTP rates from the Utility-1 model.
Figures 14, 15, and 16 give comparisons between optimal and near-optimal results for state of
charge, chiller cooling loads, and chiller supply temperature setpoint. For this particular exam-
ple, the near-optimal strategy nearly reproduces the optimal discharge strategy and makes full
use of storage during the period having the highest utility rates. During the period when the
chiller is off, the chiller supply temperature is simply the load return temperature. At other times
during the occupied period, chiller priority is employed with a chiller supply temperature set-
point of 40F (4.4C). There is barely enough time during the unoccupied period to fully charge
storage, so there are no differences between the two charging strategies. The chiller operates at
the minimum chiller supply setpoint throughout the charging period. Differences in charging
strategy would be expected for days having lower loads and for plants that operate more effi-
ciently at part-load conditions.
Figure 17 shows comparisons between optimal and near-optimal states of charge for an office
building located in Houston during the first week in July beginning on a Friday. The trajectories of
states for both optimal and near-optimal control strategy are nearly identical for days where stor-
age is completely utilized. On other days, the near-optimal strategy uses less storage, performs
charging at later times, and performs discharging at earlier times compared to optimal control.
However, the impact of these differences on daily operating costs is relatively small as shown in
Figure 18. Similar results were obtained for other times of year and other systems and locations. In
general, the near-optimal control method performs very well in relation to optimal control.
The simplified controller was evaluated extensively through comparison with optimal control
for a range of different systems and locations. Simulations were performed for seven-month
Q

load, k
x
k
off
x
k
I
Q

load, k
t
k

Cap
s
------------------------- .
k k
I
=
k
off
1

=
x
k
off
x
min
>
VOLUME 13, NUMBER 4, JULY 2007 575
Figure 14. RTP rates and storage states for simple example.
Figure 15. Building and chiller cooling loads for simple example.
576 HVAC&R RESEARCH
Figure 16. Wet-bulb and chiller supply temperatures for simple example.
Figure 17. Example comparison of optimal and near-optimal storage states (office in
Houston, single-stage water-cooled chiller, minimum chiller sizing, Utility-1 rates, start-
ing at 6 a.m. on July 1).
VOLUME 13, NUMBER 4, JULY 2007 577
periods from April 1 to October 30 for the evaluation matrix presented in Table 2. All combina-
tions of these variables were considered, resulting in 576 different seven-month simulations. All
total, this required 123,264 daily optimizations for the benchmark optimization. Perfect knowl-
edge of utility rates and loads was assumed for both the optimal and near-optimal control meth-
ods in all cases.
Figure 19 gives comparisons of costs per unit of cooling over the cooling season ($/ton-h)
between optimal and near-optimal control for all combinations of parameters in Table 2. The
near-optimal costs are generally within about 2% of the minimum costs. It was found that the
agreement is slightly better for the two plants having no dependence of COP on partial loading.
Figure 18. Example comparison of optimal and near-optimal daily costs (office in Hous-
ton, single-stage water-cooled chiller, minimum chiller sizing, Utility-1 rates, starting at
6 a.m. on July 1).
Table 2. Controller Evaluation Matrix
Parameter Range of Values
Building type Office, School, Retail Store, Warehouse
RTP pricing Utility-1 and Utility-2 Rate Models
Locations Chicago, Houston, New York, San Francisco
Storage size 50%, 100%, 200% of Storage Size for Minimum Size Chiller
Storage type Internal Melt, Ice-on-Pipe
Chiller type
Water-Cooled Centrifugal PlantSingle Stage
Water-Cooled Centrifugal PlantInfinite Staging
Air-Cooled Screw Chiller
Chiller size 10% and 30% Oversizing Based on Design Day
578 HVAC&R RESEARCH
Furthermore, the largest relative errors occur on days with low loads and low costs. The impacts
of these differences on annual cost differences are very small.
SUMMARY
The control strategy described in this paper for cold storage systems having real-time pricing
(RTP) utility rates can be implemented at relatively low cost and performs very well compared
to optimal control. The algorithm requires relatively low-cost measurements, requires very little
plant-specific information, is computationally simple, and ensures that building cooling require-
ments are always met (e.g., storage isnt prematurely depleted). The most common control strat-
egy employed for cold storage systems is chiller-priority control. However, as shown by Braun
(2007), the penalty associated with employing chiller-priority control is very significant for sys-
tems having RTP rates. In fact, the use of storage with chiller-priority control can actually result
in higher costs than without storage.
ACKNOWLEDGMENTS
The financial support of ASHRAE under RP-1252 and the technical support provided by the
Project Monitoring Subcommittee that included Steve Blanc, John Dunlap, Michael Kint-
ner-Meyer, Robert Sonderegger, and Gene Strehlow are greatly appreciated. The RTP model
and cooling load data were developed by Changlin Sun under the supervision of Keith Temple
and Todd Rossi at Field Diagnostics Services, Inc.
NOMENCLATURE
Cap
s
= storage capacity defined as the maxi-
mum change in internal energy of the
storage tank that can occur during a
discharge cycle
c
f
= specific heat of the water/glycol mix-
ture flowing through chiller and stor-
age
Figure 19. Optimal vs. near-optimal cooling season costs per unit of cooling (all combina-
tions of Table 2 parameters).
VOLUME 13, NUMBER 4, JULY 2007 579
COP = cooling plant coefficient of perfor-
mance defined as the ratio of the
plant cooling load to the power con-
sumption
COP
c
= representative cooling plant COP
associated with charging storage
COP
d
= representative cooling plant COP
associated with discharging of stor-
age under a chiller-priority strategy
k = stage within algorithms for optimiza-
tion or simplified control strategy
J = utility cost for the optimization
period
= mass flow rate of water/glycol mix-
ture flowing through chiller
N = number of time stages in the optimi-
zation period
P = total plant power consumption
R = cost per unit of electrical energy
RTP
th
= threshold RTP rate for economic dis-
charge of storage
RTP
c
= representative RTP rate for the charg-
ing period
= chiller cooling load
= total plant cooling load (building
load)
T
amb
= ambient dry-bulb temperature
T
coil
= temperature of water/glycol mixture
supplied to load
T
chws
= temperature of the water/glycol mix-
ture supplied by the chiller
T
max
= maximum ambient dry-bulb tempera-
ture for the day
T
s
= temperature at which the storage
medium melts or freezes (32F [0C]
for this study)
T
wb
= ambient wet-bulb temperature
u = rate of energy addition to storage
over the stage (positive for charging
and negative for discharging)
u
min
= maximum storage discharge rate
(minimum charging rate)
u
max
= maximum storage charge rate
x = relative state of charge for storage
(0 for 100% water and 1 for
100% ice)
t = simulation time step (stage interval)

c,k,max
= maximum heat transfer effectiveness
for charging at the current stage k if
all of the chilled-water/glycol flow
passes through the tank (no bypass)

d,k,max
= maximum heat transfer effectiveness
for discharging at the current stage k
if all of the chilled-water/glycol flow
passes through the tank (no bypass)
Subscripts
c = charging condition
d = discharging condition
k = stage
max = maximum possible
min = minimum possible
on = effective on-peak period
off = off-peak rates
REFERENCES
ASHRAE. 2003. 2003 ASHRAE HandbookHVAC Applications, Chapter 41, Supervisory Control Strat-
egies and Optimization, pp. 41.141.39. Atlanta: American Society of Heating, Refrigerating and
Air-Conditioning Engineers, Inc.
Braun, J.E. 1992. A comparison of chiller-priority, storage-priority, and optimal control of an ice-storage
system. ASHRAE Transactions 98(1):893902.
Braun, J.E. 2007. Impact of control on operating costs for cool storage systems with dynamic electric rates.
ASHRAE Transactions 113(2).
Drees, K.H., and J.E. Braun. 1995. Modeling of area constrained ice storage systems. HVAC&R Research
1(2):14359.
Drees, K.H., and J.E. Braun. 1996. Development and evaluation of a rule-based control strategy for ice
storage systems. HVAC&R Research 2(4):31236.
Henze, G.P., R.H. Dodier, and M. Krarti. 1997. Development of a predictive optimal controller for thermal
energy storage systems. HVAC&R Research 3(3):23364.
Henze, G.P., and M. Krarti. 1999. The impact of forecasting uncertainty on the performance of a predictive
optimal controller for thermal energy storage systems. ASHRAE Transactions 105(2):55361.
Henze, G.P. 2003a. An overview of optimal control for central cooling plants with ice thermal energy stor-
age. Journal of Solar Energy Engineering 125(3):302309.
m

chw
Q

ch
Q

load
580 HVAC&R RESEARCH
Henze, G.P. 2003b. Impact of real-time pricing rate uncertainty on the annual performance of cool storage
systems. Energy and Buildings 35(3):31325.
NREL. 1995. Users Manual for TMY2s (Typical Meteorological Years), NREL/SP-463-7668, and TMY2s,
Typical Meteorological Years Derived from the 19611990 National Solar Radiation Data Base,
CD-ROM. Golden, CO: National Renewable Energy Laboratory.
Rawlings, L.K. 1985. Strategies to optimize ice storage. ASHRAE Journal 27:3944.
Spethmann, D.H. 1989. Optimal control for cool storage. ASHRAE Transactions 95(1):118993.
Sun, C., K. Temple, T. Rossi, and J.E. Braun. 2006. Interaction between dynamic electric rates and thermal
energy storage control, Final Report for RP-1252, American Society of Heating, Refrigerating and
Air-Conditioning Engineers, Inc., Atlanta, GA.
Tamblyn, R.T. 1985. Control concepts for thermal storage. ASHRAE Transactions, 91(1B):511.
West, J.D., and J.E. Braun, 1998. Modeling partial charging and discharging of area-constrained ice stor-
age tanks. HVAC&R Research 5(3):20928.

También podría gustarte