See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/261281692

Extremum seeking control of cooling tower for self-optimizing efficient

operation of chilled water systems

Conference Paper  in  Proceedings of the American Control Conference · June 2012

DOI: 10.1109/ACC.2012.6315202

CITATIONS READS

18 89,127

4 authors:

Xiao Li Yaoyu Li
ANSYS University of Texas at Dallas
13 PUBLICATIONS   235 CITATIONS    163 PUBLICATIONS   3,729 CITATIONS   

SEE PROFILE SEE PROFILE

John E. Seem Pengfei Li

Private Investor United Technologies Research Center
68 PUBLICATIONS   1,805 CITATIONS    18 PUBLICATIONS   345 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Renewable Energy Optimization View project

President of High Altitude Trading, Inc. View project

All content following this page was uploaded by John E. Seem on 30 June 2015.

The user has requested enhancement of the downloaded file.

2012 American Control Conference
Fairmont Queen Elizabeth, Montréal, Canada
June 27-June 29, 2012

Extremum Seeking Control of Cooling Tower for Self-optimizing

Efficient Operation of Chilled Water Systems
Xiao Li, Yaoyu Li, Member, IEEE, and John E. Seem, Member, IEEE, Pengfei Li

Abstract—This paper presents an extremum seeking control
(ESC) scheme for energy efficient operation of chilled-water
system, and Modelica based dynamic simulation models for
demonstrating the effectiveness of the proposed control strategy.
The simulated plant consists of a water-cooled screw chiller and
a mechanical -draft counter-flow wet cooling tower. The ESC
scheme takes the total power consumption of chiller compressor
and tower fan as feedback, and uses the fan speed setting as
control input. The inner-loop controllers for the chiller
operation include two PI control loops for regulating the
evaporator superheat and the chilled water temperature.
Simulation was conducted for different scenarios. The
simulation results demonstrated the effectiveness of the
proposed ESC strategy, and the potential for energy saving is
also evaluated. Also investigated is the ESC windup issue caused
by the limitation of the fan’s maximum operation speed. A
back-calculation anti-windup ESC scheme is applied and its
effectiveness is validated by simulation results.
I. INTRODUCTION
consumption shows a convex characteristic with a global
minimum (Fig. 2), which facilitates gradient search type of
optimization. The authors proposed an open-loop control
scheme to search for the nearly optimal fractional tower
airflow based on the parameter estimation of the design
characteristics of the chiller and cooling tower. Similar to this
study, most existing methods to the control and optimization
of the chilled-water systems have been based on
nominal/empirical models. In practice, due to the unknown
environment changes and the hard estimated system
degradation, such models may often be inaccurate. Therefore,
real-time optimization of set-point tracking without exact
system knowledge is more desirable for operations of the
chiller plant. The extremum seeking control (ESC), as a major
class of self-optimizing strategies, has recently drawn
significant attention for HVAC applications.

H VAC systems account for about 32% of the energy used

in the commercial buildings [1]. Figure 1 shows the
schematic of typical chilled-water ventilation and
air-conditioning system for commercial buildings with three
main components: air handling unit, chiller and cooling
tower. The power consumption of such system is mainly from
the chiller compressor and the cooling tower fan. Due to
significant variations in ambient, load and equipment
conditions, developing proper control strategy is critical for
efficient operation of chilled-water systems.
Modeling, control and optimization techniques have been Fig. 1. Schematic of a typical chilled-water system.
well investigated for the chilled-water system [1-5], e.g. the The objective of ESC is to search for the optimal input in
supervisory control strategy by Kaya and Sommer [3] and the real time in a nearly model-free fashion [6]. Li et al. [7]
hourly regulated optimal control scheme by Liu and Chuah presented an ESC scheme for efficient operation of the
[4]. Braun and Diderrich [2] studied the coupling between the air-side economizer. For chilled-water systems, application
power consumption of chiller compressor and cooling tower of ESC has been recently investigated [1, 5]. Tyagi et al. [5]
fan. As shown in Fig. 2, they demonstrated that the tower fan presented a work with golden-section search as their ESC
power increases with the relative tower airflow, while the solution for determining the optimal condenser supply water
chiller power decreases. As net effect, the total power temperature based on an oracle function. Such scheme may
lead to long searching time in practice since every step of
Manuscript received March 15, 2012. This work was supported by the search needs to wait for the system transient to settle. For the
Johnson Controls, Inc. same problem, Sane et al. [1] described a dither ESC solution,
Xiao Li is with University of Wisconsin - Milwaukee, 3200 N. Cramer St., where again the condenser supply temperature is used as the
Milwaukee, WI 53211, USA (corresponding author; tel: (414) 426-7497;
e-mail: xiaoli@uwm.edu). control input. Simulation results were shown without
Yaoyu Li is with University of Texas at Dallas, 800 W. Campbell Rd., mentioning the details about the simulation platform. As the
Richardson, TX 75080, USA (e-mail: yaoyu.li@utdallas.edu). dither ESC is a dynamic scheme of gradient search, a key
John E. Seem is with Building Efficiency Research Group, Johnson aspect of its design is to compensate for the input and output
Controls, Inc., 507 E. Michigan St., Milwaukee, WI 53202, USA (e-mail:
john.seem@gmail.com).
dynamics. An ESC design of such may not be sufficient
Pengfei Li is with United Technologies Research Center, East Hartford, without simulation on a dynamic model of cooling tower and
CT, 06108, USA (e-mail: lip1@utrc.utc.com).

978-1-4577-1094-0/12/$26.00 ©2012 AACC 3396

chiller. Also, the control input adopted in [1] and [5], i.e. the
condenser supply water temperature, is not a variable that can
be directly manipulated in practice. Some inner loop control
of the cooling tower must be implemented. In addition, its
operation contains another uncertainty, i.e. the reliability of
the associated temperature sensor.
Fig. 2. Tradeoff between energy consumption of chiller and cooling tower.
Fig. 3. Basic system diagram of ESC working on Chiller-Tower ssystem.
With these concerns considered, this study approaches to
the ESC based chilled-water system control problem with two
different perspectives. First, a Modelica [8] based dynamic
simulation model platform is developed for the cooling tower
in Dymola [9]. Second, the cooling tower fan speed, instead
of the condenser supply water temperature, is used as the
input for the ESC control design. With the typical
variable-speed drive (VSD) equipped for the cooling towers
nowadays, setting the VSD frequency or the motor speed is
direct and simple. No cost or uncertainty is needed for the
temperature sensor than otherwise. With constant condenser
water flow rate assumed, the total power consumption of the
chiller compressor and the cooling tower fan is used as the
performance index for feedback. In particular, a screw chiller
is chosen for the simulation plant model, which would not
affect the generality of the results for plants with other types
of chillers. The proposed ESC framework is illustrated in the
schematic in Fig. 3.
In practical chiller operation, the tower fan has both
maximum and minimum speed limits. The control input
(cooling tower fan speed) may be saturated to its upper bound
during certain operating conditions. Due to the presence of
3397
integrator in the ESC loop, integral windup is inevitable for
ESC system operations and may slow down or disable the
optimal searching ability of ESC. Li and Seem [10] proposed
a back-calculation based anti-windup ESC scheme, which
was later applied by [7] for the ESC operation of the air-side
economizer. In this study, this scheme is applied to the ESC
operation of chilled water system under fan speed saturation.
II. DYNAMIC MODELING OF CHILLER-TOWER SYSTEM
The dynamic model of chiller-tower system is developed
with Dymola 2012 FD1 [9], TIL Library 2.0.1 and TIL Media
Library 2.0.4 [11]. The cooling tower model follows our
recent work in [12], which is a modification of the work by
[13]. The screw compressor modeling follows the work by
[14], while the condenser, evaporator and expansion device
models are based on the work of [15]. For other components,
such as fan, pump, collection basin and valve, we have
adopted those available in the TIL Library 2.0.1. Two control
loops have been implemented for the screw chiller model: one
for the evaporator superheat control and the other for the
chilled water temperature regulation. All the controllers and
plant model are integrated into simulation platform.
A. Dynamic Modeling of Cooling Tower System
Cooling towers reject heat via evaporative cooling [16]. Hot
water is sprayed from the top of the cooling tower, and air is
drawn from the bottom by the tower fan. As water falls
through the fill, water temperature decreases due to
evaporative cooling. In this study, the finite volume method is
used to model the one-dimensional heat and mass transfer
process. The dynamic balances of mass and energy are
established for both the water and air sides. The transient
mass and energy storage is treated at the water side but
neglected at the air side.
Following [12], for the ith water-side control volume, the
energy balance is established as
dT
mw,i  cp,w,i  w,i  m w,in,i  hw,in,i  hw,i   m w,out ,i  hw,out ,i  hw,i   qi (1)
dt
where mw,i is the mass of water stored in the cell, cp,w,i is the
specific heat of water (which can be determined by the local
water temperature Tw,i), ṁw,in,i and ṁw,out,i are the mass flow
rates for the inlet and outlet water flow, respectively, hw,in,i
and hw,out,i are the specific enthalpy of the inlet and outlet
water flow, respectively, and hw,i is the specific enthalpy of
water in the cell. qi is the heat flow transferred to the
neighbored (also the ith) moist air cell, which includes both
sensible heat flow and latent heat flow due to evaporation.
For the mass balance of the same water-side control
volume, cell volume Vcell is assumed as constant, while water
density ρw,i may change with evaporation and temperature
change in the cell. The dynamic mass balance of the ith water
cell is given by [12]:
dT V dm w,in
m w,in,i  m w,out ,i  m evap,i  Veffective  w,i w,i w,i  cell (2)
dt AT vw dt
where ṁevap,i is the vapor mass transfer flow rate into the
moist air side. Veffective is the water droplet volume in the cell.
The ratio of water droplet per unit volume of the tower is
around the level of 0.001 [17]. vw is the velocity of water
droplets under free fall (no packing), is assumed constant. AT
is the cross-sectional area of the tower. β= -(əρ/ əT)P/ρ is the
isobaric coefficient of expansion. βw,i and ρw,i can be
determined by the local water temperature.
At the air side, the steady-state relations were derived
following the detailed analysis model [13], i.e.
H a ,in ,i  H a ,out ,i  q sen ,i  qlat ,i  0 (3)
The sensible and latent heat flow rates can be determined by
qsen ,i  hC ,i AV Vcell Tw,i  Ta ,i  (4)
qlat ,i  h f , g ,i  m evap ,i  h f , g ,i  hD ,i AV Vcell s , w,i  a ,i 
where hC,i is the local heat transfer coefficient, AV is the
surface area of water droplets per unit volume, Ta,i is the local
air temperature, hf,g,i is the latent heat of vaporization
depending on the local water temperature. hD,i is the local
mass transfer coefficient, ωs,w,i is the saturated-air humidity
ratio at the local water temperature, and ωa,i is the local
humidity ratio of the moist air.
The fill is used in most cooling towers, however, it is
usually hard to predict its heat rejection performance
analytically because of the difficulty in evaluating the contact
time and the surface area between the air and the water
through the fill [18]. Due to the difficulty in getting a general
correlations for heat and mass transfer in cooling tower in
terms of the physical tower characteristics, the NTU and the
Lewis relation Lef have been used to characterize the heat and
mass transfer coefficients for specific tower designs [13].
The mass transfer coefficient can be derived by using the
overall NTU for mass transfer, i.e.
NTU  m a ,in
hD ,i AV  (6)
VT
where VT is the total tower volume and ṁa,in is the inlet-air
flow rate and it varies with the tower geometry, NTU and air
inlet flow rate. The heat transfer coefficient is determined by
Le f  NTU  c pm ,i  m a ,in
hC ,i AV  (7)
VT
and the local specific heat of moist air cpm,i is determined by
c pm ,i  c pa ,i  a ,i c pv ,i , where cpa,i is the local specific heat of
dry air and cpv,i is the local specific heat of water vapor [19].
hC,i may change with the local value of Le f and c pm ,i . NTU
could be given with empirical parameters
NTU  c  m w,in / m a ,in 
n 1
, where c and n are empirical
constants specific to a particular tower design. They can be
fitted from the performance measurements for a specific
tower on a log-log plot [19].
Kloppers and Kröger [20] suggested that the equation by
[21] should be used for getting the value of Lewis relation,
which is given by:
   d     d 
Le f  Le 2/3  s , w  1 / ln  s , w  (8)
 a  d   a  d  
and a numerical value of 0.92 be preferred when the fill
performance test data is insufficient to accurately predict the
Lewis relation of a particular fill. The cooling tower fan
model follows TIL.MoistAirComponents.Fans.Fan2ndOrder
in the TIL Library [11]. The detailed description of the model
could be find in [12].
B. Modeling and Inner-loop Controls of Screw Chiller
1) Screw Compressor
The dynamics of screw compressor is much faster than that of
the entire chiller cycle. For this study, a transient model of
screw compressor is not necessary. Instead, the polytropic
static compression model developed [14] is adopted
m c ,in  scomp c ,in ncVc , maxc , v (9)
(5)
c
 
Pc ,out  Pc ,in  c , out  (10)
  
 c ,in 
where ṁc,in is the refrigerant mass flow rate at the compressor
inlet, and scomp[0, 1] is the slide-valve control, which
determines the compressor load. scomp is used to regulate the
chilled water temperature. ρc,in and ρc,out are the refrigerant
densities at the compressor inlet and outlet, respectively. nc is
the compressor speed. Vc,max is the theoretical compressor
volume with full load condition. Pc,in and Pc,out are the
compressor inlet and outlet pressures, respectively. γc is the
ratio of specific heats in the compressor, and ηc,v is the
volumetric efficiency. The volumetric efficiency could be
obtained from the pressure volume curve [22]. ASHARE [23]
suggests linear pressure-volume characteristic for the
pressure ratio ranging from 2 to 9, for both the twin-screw
and the single screw compressors. Following [24], the
volumetric efficiency is determined as
P 
c ,v  0.95  0.0125   c ,out  (11)
 Pc ,in 
The electrical power consumed by the compressor is
m c ,out   hc , out  hc ,in 
Wc , elec  (12)
c , ac , moc , me
where m c ,out is the refrigerant mass flow rate at the
compressor outlet. hc,in and hc,out are the inlet and outlet
specific enthalpies, respectively. c , a , c , mo and c , me are the
adiabatic efficiency, motor efficiency and mechanical
efficiency, chosen as 0.8, 0.85 and 0.95, respectively [24].
2) Condenser, Evaporator and Expansion Valve
[16]: The shell-and-tube heat exchanger models of condenser
and evaporator in the screw chiller follow the work by [15].
Both exchangers are of the counter-flow type with in-tube
water flow. The finite volume method is applied for both the
water and the refrigerant sides. The heat transfer is
considered between each pair of the cells. The mass and
energy balances are set for each cell (both refrigerant and
water side).
The thermal expansion valve (TXV) model in [25] is
adopted, which assumes a quadratic relation between the
mass flow rate and the pressure difference across the valve.
The effective flow area is set as an external input, which can
be used to regulate the evaporator superheat for the chiller.
3398
No transient mass storage is assumed for the TXV, and an
isenthalpic throttling process is considered.
3) Inner Loop PI Controls
The superheat control is important for chiller operation. In
the evaporator, liquid refrigerant has a much higher heat
transfer coefficient than the vaporized refrigerant. The
two-phase flow is desired to maximize the cooling capacity of
the system. Meanwhile, if not all the liquid refrigerant gets
vaporized before entering the compressor, the remained
liquid will damage the compressor. So the evaporator
superheat must be maintained to ensure both the safety and
the efficiency [26]. Also, as required by the air handling unit
operation, the chilled water temperature is expected to be
maintained around a setpoint [23]. To satisfy these two needs,
for the chiller-tower plant simulated, two inner-loop PI
controllers are implemented to regulate superheat and chilled
water temperature by tuning the thermal expansion valve’s
effective flow area (ATXV) and the compressor slide-valve
opening (scomp), respectively. The schematic of the
chiller-tower system with both ESC controller and inner loop
PI controllers is shown in Fig. 4. The set points of the
evaporator super heat and chilled water temperature are
assumed as 5oC and 7oC, respectively.
Fig. 4. Chiller inner loop controls of superheat and chilled water temperature.
III. BRIEF REVIEW OF EXTREMUM SEEKING CONTROL
The ESC is to find the optimal input uopt(t) in real time for a
generally unknown and/or time-varying objective function
l(t,u) with online measurement of the objective value to be
minimized. Typical ESC designs are based on the assumption
of quadratic objection function, which holds for a reasonable
neighborhood of the optimum for a general setting of convex
nonlinear mapping. A typical dither ESC block diagram is
shown in Fig. 5 [27]. The transfer function FI(s) and FO(s)
denote the input dynamics and sensor dynamics, respectively.
The output of the objective function y, which may be
corrupted by noise n, is assumed to be directly observable for
feedback. The dithered output signal (by dither signal d2(t))
passes through the high-pass filter FHP(s), multiplied by the
demodulating signal d1(t) and low-pass filtered by FLP(s),
resulting in a signal proportional to gradient l / u  uˆ  . An
integrator will help to eliminate the gradient when closing the
loop. The compensator K(s) can be designed to enhance the
3399
transient performance or stability.
Fig. 5. Block diagram of dither ESC.
The dither amplitude a needs to be chosen such that it is
small enough to avoid large output oscillation, but large
enough to overcome noise effect. The dither frequency ωd
could be determined from the estimation of the system input
dynamics FI(s). FHP(s) and FLP(s) need to be designed to
locate ωd in the pass band of FHP(s) and the stop band of
FLP(s). More design details follows [29]. The ESC method
achieves the convergence to the system optimality based on
an integral action on the gradient proportional signal
extracted by the pair of dither- demodulation signals,
high-pass and low-pass filters. For practical systems, all
actuators have physical limitations, which saturate the control
actions at certain point. Therefore integral windup could be a
problem for ESC system operations. To avoid the integral
windup, a number of anti-windup techniques have been
proposed in the past [28, 29]. Li and Seem [10] proposed a
back calculation based anti-windup strategy, as shown in Fig.
6, which is compatible with the simple nature of extremum
seeking control. Recently, Li et al. [7] applied this scheme to
the air-side economizer for dealing with the damper
saturation situations. In this study, this scheme is again
applied to deal with the possible integral windup due to the
fan speed saturation.
Fig. 6. Block diagram of modified anti-windup ESC.
IV. SIMULATION
In this study, the dither ESC framework is applied to the
chiller-tower system to minimize the power consumption of
the screw chiller and cooling tower fan by tuning the tower
fan speed. The input dynamics from the tower fan speed to the
total power consumption is estimated based on the system
responses of several step changes at both sides of the
optimum, as shown in Fig. 7. The input dynamics is estimated
0.03162
FˆI  s   2 (13)
s  2 1.06  0.0316 s  0.0316 2
based on the slowest step response to achieve better
robustness, which has cutoff frequency 0.0187 rad/s. The
dither frequency d is selected as 0.0043 rad/s, and the 2nd-
order high-pass and low-pass filters are chosen to have
cut-off frequencies of 0.0025 and 0.003 rad/s, respectively.
The dither amplitude is selected as 7.1 Hz for the VSD input.
The dither phase angle is selected as -0.604 radian.
Fig. 7. Step responses of full simulation model and the 2nd-order estimate.
The designed ESC is simulated on the dynamic
simulation model of the chiller-tower system described
earlier. The ESC performance is first tested under a fixed
operating condition. The RH and temperature for the
cooling-tower inlet air flow are set as 20% and 310 K,
respectively. The temperature and mass flow rate of the
evaporator inlet water are set as 285.15 K and 13.2 kg/s,
respectively. The static map from cooling tower fan speed to
the power consumptions of the chiller compressor and the
tower fan (Fig. 8) gives the optimal fan speed and power
consumption at 250.351 Hz and 231174 W, respectively.
Fig. 8. Static map from cooling tower fan speed to total power consumption.
Fig. 9. Fan speed and power consumption for ESC with fixed condition.
The simulation first starts at a fixed fan speed of 200 Hz,
and the ESC controller is turned on at t = 5000 sec. (Fig. 9).
The ESC search results in the average steady-state fan speed
of 256.302 Hz and the total power of 232018 W, respectively,
with the 1% settling time of about 11720 sec. Compared to
the estimated optimum in the static map, the steady-state error
3400
is about 2.37% and 0.37% for the fan speed and the total
power, respectively. Figure 9 shows that the estimated
optimal fan speed falls within the range of the input dither.
The ESC controller is then tested with a ramp change in the
evaporator inlet water temperature TEW (e.g. due to a load
reduction) from 12 C to 10 C in 3000 seconds starting from
t = 60000 second. The static maps of the two conditions are
shown in Fig. 10, with the optimal point being (250.351 Hz,
231174 W) for the first condition, and (179.345 Hz , 86527.3
W) for the second condition, respectively. The first condition
is the same as the previous case of fixed condition.
In Fig. 11, the ESC searched average steady-state fan speed
and power consumption of the second condition is about
182.714 Hz and 86538.4 W, respectively, differing from the
estimated optimum by only 1.88% and 0.013%, respectively.
The power output settles within ±1% of its steady state in
about 13735 second. Also, as marked in Fig. 10, if the fan
speed remained unchanged during the ramp change, the
system would operate at point A, which consumes 91759.1W.
Therefore, ESC adapts the system operation with power
saving of 5231.8W (5.7%).
Fig. 10. Static maps for TEW decreased from 12oC to 10oC.
Fig. 11. ESC simulation results for TEW decreased from 12oC to 10oC.
The effectiveness of the back-calculation anti-windup ESC
is verified by simulation. The maximum fan speed is set at
290 Hz. The plant is assumed to experience a change in the
evaporator inlet water temperature change (load change). No
ambient environment change is applied, and all other settings
are same as the case of fixed operation condition. The
evaporator inlet water temperature is initially set at 12 C. At
60000 seconds, TEW ramps up to 13 C in 5000 seconds. At
125000 sec, another 5000 sec ramp change brings TEW back to
 12oC. Figure 12 shows the integral windup problem of the
standard ESC with actuator saturation. In the upper subplot,
the optimum is successfully achieved by ESC search before
the first ramp change. From 60000 sec, under the change of
system condition, the fan speed is saturated at 290 Hz. When
the second ramp change is applied, the ESC fails to search for
the new optimum due to the integral windup. The fan speed is
saturated at the 290 Hz limit.
The back-calculation based anti-windup ESC scheme is
applied to the same case. In Fig. 13, the anti-windup ESC
responds to the system condition change when the second
ramp change of TEW starts and converges to the new
optimum successfully. The simulation results show that the
modified anti-windup ESC scheme could effectively solve
the potential integral windup problem caused by system
actuator saturation.
Fig. 12. Integral windup: standard ESC with actuator saturation.
Fig. 13. Antiwindup: modified antiwindup ESC with actuator saturation.
V. CONCLUSION
This paper presents an ESC based cooling tower control
scheme which can minimize the combined power
consumption of cooling tower fan and chiller compressor.
The ESC strategy is tested on a dynamic simulation model of
the chiller-tower system developed in Dymola using
Modelica. The inner loop controls of superheat and chilled
water temperature are implemented, by regulating the valve
flow area and slide-valve opening, respectively. Simulation
study was performed for a scenario of fixed condition and
then for two scenarios of varying conditions in which ramp
changes are introduced to the evaporator inlet water
temperature and the ambient air condition, respectively. The
ESC searched results show very small steady-state errors
compared to the pre-calibrated static maps, with reasonable
settling time even under varying conditions. The power
saving performance is also evaluated for the simulated
examples. The effectiveness of the back-calculation based
anti-windup ESC is also validated by simulation.
REFERENCES
[1] H. S. Sane, C. Haugstetter and S.A. Bortoff, "Building HVAC Control
Systems - Role of Controls and Optimization," in Proceedings of ACC,
Minneapolis, Minnesota, 2006, pp. 1121-1126.
[2] J. E. Braun and G. T. Diderrich, "Near-Optimal Control of Cooling
Towers for Chilled Water Systems," ASHRAE Transactions vol. 96, pp.
806-816, 1990.
3401
View publication stats
[3] A. Kaya and A. Sommer, "Energy Management of Chillers by
Multilevel Control and Optimization," Journal of Dynamic Systems,
Measurement, and Control vol. 107, pp. 270-277, 1985.
[4] C.-W. Liu and Y.-K. Chuah, "A Study on an Optimal Approach
Temperature Control Strategy of Condensing Water Temperature for
Energy Saving," Int. J. of Refrig, vol. 34, pp. 816-823, 2011.
[5] V. Tyagi, H. Sane and S. Darbha, "An Extremum Seeking Algorithm
for Determining the Set Point Temperature for Condensed Water in a
Cooling Tower," in Proceedings of American Control Conference,
Minneapolis, Minnesota, 2006, pp. 1127-1131.
[6] K. B. Ariyur and M. Krstić, Real-Time Optimization by
Extremum-Seeking Control. New York: John Wiley & Sons, Inc., 2003.
[7] P. Li, Y. Li and J.E. Seem, "Efficient Operation of Air-Side Economizer
Using Extremum Seeking Control," Journal of Dynamic Systems,
Measurement and Control, vol. 132, pp. 031009_1-031009_102010.
[8] Modelica Association, Modelica and Modelica Association, 2012
Available: http://www.modelica.org/
[9] Dassault Systèmes, Dymola: Multi-Engineering Modeling and
Simulation, 2012. Available:
http://www.3ds.com/products/catia/portfolio/dymola
[10] Y. Li and J.E. Seem, "Extremum Seeking Control with Actuator
Saturation Control," US Patent 20100106331, 2010.
[11] TLK-Thermo GmbH, Engineering Services and Software for Thermal
Systems, 2012. Available: http://www.tlk-thermo.com/index.php
[12] X. Li, Y. Li and J.E. Seem, "Dynamic Modeling of Mechanical Draft
Counter-Flow Wet Cooling Tower with Modelica," in Proceedings of
the 4th National Conf. IBPSA-USA, New York, 2010, pp. 193-200.
[13] J.E. Braun, S.A. Klein and J.W. Michell, "Effectiveness Models for
Cooling Towers and Cooling Coils," ASHRAE Transactions, vol. 95,
pp. 164-174, 1989.
[14] W. Zhang, C. Zhang and G. Ding, "Transient Modeling of an
Air-Cooled Chiller with Economized Compressor. Part I: Model
Development and Validation," Appl. Therm. Eng, vol. 29, pp.
2396-2402, 2009.
[15] P. Li, Y. Li and J.E. Seem, "Consistent Initialazation of System of
Differential-Algebraic Equations for Dynamic Simulation of
Centrifugal Chillers," Journal of Building Performance Simulation,
vol. 5, pp. 115-139, 2012.
[16] D. G. Kröger, Air-Cooled Heat Exchangers and Cooling Towers -
Thermal Flow Performance Evaluation and Design. Tulsa, OK:
PennWell Corporation, 2004.
[17] M. A. Bernier, "Thermal Performance of Cooling towers," ASHRAE
Journal vol. 37, pp. 56-61, 1995.
[18] M. A. Bernier, "Cooling Tower Performance: Theory and
Experiments," ASHRAE Transactions, vol. 100, pp. 114-121, 1994.
[19] J. E. Braun, "Methodologies for the Design and Control of Central
Cooling Plants," Department of Mechanical Engineering University of
Wisconsin - Madison, Ph. D. Thesis, 1988.
[20] J. C. Kloppers and D. G. Kröger, "The Lewis Factor and Its Influence
on the Performance Prediction of Wet-Cooling Towers," International
Journal of Thermal Science, vol. 44, pp. 879-884, 2005.
[21] F. Bošnjaković, Technische Thermodinamik, Theodor Steinkopf,
Dresden, 1965.
[22] P. C. Hanlon, Compressor Handbook. New York: McGraw-Hill, 2001.
[23] ASHRAE, ASHARE Handbook - HVAC Systems and Equipment (SI).
Atlanta, GA.: American Society of Heating, Refrigerating, and
Air-Conditioning Engineers Inc. , 2008.
[24] L. Fu, G. Ding, Z. Su and G. Zhao, "Steady-State Simulation of Screw
Liquid Chillers," Appl. Therm. Eng, vol. 22, pp. 1731-1748, 2002.
[25] C. C. Richter, "Proposal of New Object-Oriented Model Libraries for
Thermodynamic Systems,", Institute for Thermodynamics, Technical
University at Braunschweig, Ph. D. Thesis, 2008.
[26] M. Elliott, Z. Walton, B. Bolding and B.P. Rasmussen, "Superheat
Control: A Hybrid Approach," HVAC&R Research, vol. 15, pp.
1021-1043, 2009.
[27] M. A. Rotea, "Analysis of Multivariable Extremum Seeking
Algorithms," in ACC, Chicago, IL, 2000, pp. 433-437.
[28] K. J. Åström and T. Hägglund, PID Controllers: Theory, Design, and
Tuning. Research Triangle Park: Instrument Society of America, 1995.
[29] P. March and M. Turner, "Anti-Windup Compensator Designs for
Nonsalient Permanent-Magnet Synchronous Motor Speed Regulators,"
IEEE Trans on Industry Applications, vol. 45, pp. 1598-1609, 2009.