Building and Environment 44 (2009) 11881198

Contents lists available at ScienceDirect

Building and Environment

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

An optimal control strategy for complex building central chilled water systems for practical and real-time applications
Zhenjun Ma, Shengwei Wang*
Department of Building Services Engineering, The Hong Kong Polytechnic University, Hung Hom, Kowloon, Hong Kong

a r t i c l e i n f o
Article history: Received 30 June 2008 Received in revised form 27 August 2008 Accepted 27 August 2008 Keywords: Chilled water system Optimal control Energy efciency Online application

a b s t r a c t
This paper presents an optimal control strategy for online control of central chilled water systems in complex building air-conditioning systems to enhance their energy efciency. The optimal control strategy is formulated using a systematic approach by considering the system level and subsystem level characteristics and interactions among the overall system. The requirements and constraints of practical applications are also carefully considered during the development of this strategy. This optimal control strategy consists of the model-based performance predictor (i.e., simplied models), cost estimator (i.e., cost function), optimization technique, supervisory strategy and a number of local control strategies. The local control strategies are used to ensure the robust operation and keep track of control settings considering the dynamic characteristics of the local process environment. The performance of this strategy is tested and evaluated in a simulated virtual environment representing the complex central chilling system in a super high-rise building by comparing it with that of other control strategies. The results showed that this strategy is more energy efcient and cost effective than the other strategies for online applications. This strategy is being implemented in the super high-rise building under study for eld application and validation. 2008 Elsevier Ltd. All rights reserved.

1. Introduction Building central chilled water systems are among the major functions in heating, ventilating and air-conditioning (HVAC) systems and they often contribute a signicant amount of the total energy consumption used in HVAC systems. Due to the increasing number of high-rise buildings built and growing concern on building energy consumption, the congurations and design philosophies of building central chilled water systems are becoming more and more sophisticated. How to achieve energy efcient control and operation of these complex central chilled water systems to minimize their energy inputs while providing the robust operational performance is one of the major challenges encountered by building professionals and building operators nowadays. Aiming at enhancing system operating efciency and providing robust and stable control performance, many researchers and experts in the HVAC eld have devoted considerable efforts on energy efcient control and operation of building central chilled water systems during the past two decades. These efforts have resulted in a number of near optimal and optimal control strategies

* Corresponding author. Tel.: 852 27665858; fax: 852 27746146. E-mail address: beswwang@polyu.edu.hk (S. Wang). 0360-1323/$ see front matter 2008 Elsevier Ltd. All rights reserved. doi:10.1016/j.buildenv.2008.08.011

readily available in literature [110]. Among the existing studies, Rishel [13] paid great efforts on the proper control and operation of variable speed pumps in chilled water distribution systems. Moore and Fisher [4] presented that the pump pressure differential set-point can be continuously optimized based on the opening signals of water control valves. To keep one control valve almost fully open at all times can minimize the pump energy. Ahmed [5] discussed the conventional methods for the control of variable water volume (VWV) systems and compared them with a direct digital control (DDC)-based VWV system. The simulated ow head curves showed that substantial operating costs can be saved thanks to the use of the DDC for hydraulic systems. For the system-based control and optimization, Braun et al. [6,7] presented two methodologies for determining the optimal control settings for chilled water systems without storage. One was a component model-based nonlinear optimization algorithm, in which the power consumptions of major components in the chilled water system were expressed as quadratic relationships. This methodology was used as a simulation tool for investigating the system performance. The other was a system-based near optimal strategy, in which the total power consumption of the overall chiller plant was expressed as a quadratic relationship. Brauns method was further extended by Cascia [8] through simplifying the component models, and the power consumptions of major components were expressed as a function of the temperature

Z. Ma, S. Wang / Building and Environment 44 (2009) 11881198

1189

difference between the chilled water supply and return temperatures. Lu et al. [9] presented a model-based optimization strategy for both chilled water systems and air systems. An adaptive neuron-fuzzy inference system was used to model the duct and pipe network and obtain the optimal pressure differential setpoint, while a modied genetic algorithm (GA) was employed to search for the optimal control settings. The simulation results based on a typical HVAC system showed that about 89% of operating costs of chilled water systems and air systems can be reduced when this optimization strategy is used. Jin et al. [10] presented three optimal control strategies for chilled water systems, i.e., the control of the supply head of secondary pumps, the control of the chilled water supply temperature set-point and the control of both in series. The simulation results based on a small scale HVAC system in the selected typical summer day and spring day showed that these three strategies can save about 3.853.90%, 2.533.38% and 1.9815.96% of the total system energy consumption, respectively, as compared with the conventional strategies using the xed temperature and/or xed pressure differential set-points. All these studies primarily demonstrated energy saving potentials in buildings, particularly in chilled water systems, associated with the use of improved control and optimization strategies. However, most of these strategies for the overall system optimization cannot satisfy the requirements of practical applications when the system operating efciency, control robustness and computational cost are of concern simultaneously. For instance, some strategies utilized empirical models to predict the system energy performance and environmental quality as well as the system response to the changes of control settings. These empirical models cannot always ensure stable performance prediction although they are simple. They are reliable only for the operating points within the range of the training data covered, and extrapolation outside this range may lead to signicant errors. Some strategies used global optimization techniques (i.e., genetic algorithm) to seek the optimal solutions. However, these global optimization techniques always require high computational costs and memory demands, which may seriously prevent their online applications [11], although they are effective. More importantly, the existing strategies were developed mainly for simple and typical chilled water systems, and efforts and reports on optimal control strategies for complex chilled water systems still seem missing. This paper therefore aims at developing an optimal control strategy for complex central chilled water systems for practical and real-time applications. The system level and subsystem level characteristics and interactions among the overall system, and the requirements and constraints of practical applications are carefully considered during the development of the optimal control strategy. The performance of this strategy is tested and evaluated in a simulated virtual environment representing the complex central chilling system in a super high-rise building in Hong Kong by comparing it with that of other control strategies. 2. Building and system description The strategy presented in this paper is developed as part of the building energy optimizer to be implemented in a super high-rise commercial building currently being constructed in Hong Kong. Therefore, the strategy presented in this paper is customized for and tested on the complex central chilled water system of that building. This building is a super high-rise building of approximately 490 m height (currently the tallest building in Hong Kong) and 321,000 m2 of oor area. The building has a basement of four oors, a block building of six oors and a tower building of 98 oors. The basement is used mainly for parking. The block building from the ground oor to the fth oor serves mainly as the commercial center including hotel ballrooms, shopping arcades

and arrival lobbies. The tower building is mainly for commercial ofces and a six-star hotel on the upper oors. The water system in this building is divided into ve zones to avoid the chilled water pipelines and terminal units from suffering extremely high pressure (i.e., the highest static pressure of over 40 bar and the designed working pressure of nearly 60 bar). The oors below the sixth oor are Zone 1. Zone 2 involves the oors from the seventh oor to 41st oor. Zone 3 is from the 43rd to 77th oor and Zone 4 is from the 79th to 98th oor. The six-star hotel located on the upper oors of the building is Zone 5. The sixth, 42nd, 78th and 99th oors are used as the mechanical oors to accommodate the mechanical equipment such as chillers, cooling towers, heat exchangers, pumps, PAUs and fans, etc. Considering the usage characteristics of the hotel, the separate air-cooled chillers located on the 99th oor are designed to provide the cooling source for Zone 5. For the other zones, the cooling source is provided by the water-cooled chillers located on the sixth oor. The schematics of the central chilled water system of these four zones are illustrated in Fig. 1. In this central chilled water system, six identical high voltage (10,000 V) centrifugal chillers with the capacity of 7230 kW each and the nominal power consumption of 1270 kW each at the full load condition are used to supply the chilled water at 5.5  C. Each chiller is associated with one constant condenser water pump and one constant primary chilled water pump. In the secondary chilled water system, only Zone 2 (indicated as B in Fig. 1) is supplied with the secondary chilled water directly. For the other three zones, the heat exchangers are used to transfer the cooling energy from low zones to high zones to avoid the high water static pressure. Zone 1 (indicated as A in Fig. 1) is supplied with the secondary chilled water through the heat exchangers located on the sixth oor, while the chilled water from chillers serves as the cooling source of the heat exchangers. Zones 3 and 4 (indicated as C in Fig. 1) are supplied with the secondary chilled water through the rst stage heat exchangers (HX-42 in Fig. 1) located on the 42nd oor. Some of the chilled water after the rst stage heat exchangers is delivered to Zone 3 by the secondary chilled water pumps (SCHWP-42-01 to 03) located on the 42nd oor. Some water is delivered to the second stage heat exchangers (HX-78 in Fig. 1) located on the 78th oor by the secondary chilled water pumps (SCHWP-42-04 to 06) located on the 42nd oor. The water system after the second stage heat exchangers is the conventional primarysecondary chilled water system. The conguration of the water piping system of this building is the reverse-return system. All pumps in this air-conditioning system are variable speed pumps except that the primary chilled water pumps dedicated to the chillers and heat exchangers in Zones 3 and 4 are constant speed pumps. The major design parameters of all pumps in this air-conditioning system are summarized in Table 1. 3. Formulation of the optimal control strategy In general, the control of chilled water systems is strongly affected by system designs. Different design congurations (i.e., the primarysecondary pumping system or variable primary pumping system, the water distribution systems with or without heat exchangers, the primarysecondary pumping systems with dedicated pumps or manifold pumps, etc.) may require different control strategies. It is very difcult and almost impossible to develop a general optimal control strategy that can be widely used for proper control of all possible congurations of chilled water systems. The optimal control strategy for complex chilled water systems presented in this paper is developed based on the actual complex chilled water system presented earlier. Since Zones 1 and 2 in the secondary chilled water system of that building cover all possible

1190

Z. Ma, S. Wang / Building and Environment 44 (2009) 11881198

A
FROM PODIUM & BASEMENT TO PODIUM & BASEMENT
(S-B)

C Secondary water circuit for Zone

3 and Zone 4

SCHWP-78-01 to 03

(S-B)

A Secondary water circuit for Zone 1 B Secondary water circuit for Zone 2

FROM OFFICE FLOORS (79-98) TO OFFICE FLOORSS (79-98)

D Primary water circuit E Chiller circuit F Cooling water circuit

FROM OFFICE FLOORS (43-77)

PCHWP-78-01 PCHWP-78-02 PCHWP-78-03

HX-78

HX-78

HX-78

TO OFFICE FLOORS (43-77) (S-B) SCHWP-42-01 to 03

PCHWP-42-05 PCHWP-42-06 PCHWP-42-07

SCHWP-42-04 to 06

HX-06

HX-06

FROM OFFICCE FLOORS(7-41) TO OFFICE FLOORS(7-41)

PCHWP-42-01 PCHWP-42-02 PCHWP-42-03 PCHWP-42-04

(S-B)

SCHWP-06-10 to 12

SCHWP-06-03 to 05

(S-B)

(S-B)

HX-42

HX-42

HX-42

HX-42

HX-42

HX-42

HX-42

SCHWP-06-01 to 02

SCHWP-06-06 to 09

D
PCHWP-06-01 PCHWP-06-02 EVAPORATOR WCC-06a-02 (7230 kW) CONDENSER CDWP-06-02 PCHWP-06-03 PCHWP-06-04 EVAPORATOR WCC-06a-04 (7230 kW) CONDENSER CDWP-06-04 PCHWP-06-05 EVAPORATOR WCC-06a-05 (7230 kW) CONDENSER CDWP-06-05 PCHWP-06-06 EVAPORATOR WCC-06a-06 (7230 kW) CONDENSER CDWP-06-06

E F

EVAPORAROR WCC-06a-01 (7230 kW) CONDENSER CDWP-06-01

EVAPORATOR WCC-06a-03 (7230 kW) CONDENSER CDWP-06-03

From cooling towers

Fig. 1. Schematic of the central chilled water system.

To cooling towers

control issues (i.e., the speed control of the pump distributing water to terminal units, the speed control of the pump distributing water to heat exchangers, the pump sequence control, the heat exchanger sequence control, the optimization of the pressure differential setpoint, etc.) of the overall secondary chilled water system of the building, the optimal control strategy presented in the following is therefore mainly focused on Zones 1 and 2 only in order to reduce the complexity of the control system. 3.1. Outline of the optimal control strategy The optimal control strategy developed is to optimize the operation of the central chilled water system using a systematic approach by considering the system level and subsystem level characteristics and interactions among the overall system. Fig. 2 illustrates the outline of this optimal control strategy, which consists of the model-based performance predictor (i.e., simplied models), cost estimator (i.e., cost function), optimization algorithm, supervisory strategy and a number of local control strategies. The settings optimized in this strategy include the chilled water supply temperature set-point (Tchws), the operating number of pumps distributing water to terminal units in Zone 1 (Npu,z1,ahx), the operating number of pumps in Zone 2 (Npu,z2), the operating number of heat exchangers in Zone 1 (Nhx), the optimal pressure differential set-points for Zone 1 (PDset,z1) and the optimal pressure differential set-point for Zone 2 (PDset,z2), as well as the number of chillers operating (Nch). Since the search range of the chilled water supply temperature set-point is relatively narrow (constrained from 5  C to 10  C in this

study), the exhaustive search method is used as the optimization tool to seek the global optimal solutions within this limited search range with 0.1 K increments. In this strategy, the performance predictor utilizes the performance models to predict the system energy performance and environmental quality as well as the system response to the changes of control settings under various operating conditions. The local control strategies are utilized to ensure the robust operation and keep track of the control settings considering the dynamic characteristics of the local process environment. The descriptions of the performance modes and local control strategies, and the denitions of the cost function and operating constraints used to formulate this optimal control strategy are presented in the following briey along with a detailed description of the optimization procedures.
Table 1 Specications of all pumps in the air-conditioning system. Pumps CDWP-06-01 to 06 PCHWP-06-01 to 06 SCHWP-06-01 to 02 SCHWP-06-03 to 05 SCHWP-06-06 to 09 SCHWP-06-10 to 12 PCHWP-42-01 to 07 SCHWP-42-01 to 03 SCHWP-42-04 to 06 PCHWP-78-01 to 03 SCHWP-78-01 to 03 N 6 6 1 2 3 2 7 2 2 3 2 Mw (L/s) 410.1 345.0 345.0 345.0 345.0 155.0 149.0 294.0 227.0 151.0 227.0 Head (m) 41.60 31.60 24.60 41.40 30.30 39.90 26.00 36.50 26.20 20.60 39.20

h
(%) 83.6 84.5 82.2 85.7 84.2 78.8 84.9 87.8 84.3 84.3 85.8

(S-B)

W (kW) 202 126 101 163 122 76.9 44.7 120 69.1 36.1 102

Wtot (kW) 1212 756 101 326 366 153.8 312.9 240 138.2 108.3 204

Z. Ma, S. Wang / Building and Environment 44 (2009) 11881198

1191

Online measurements and control signals Measurement filter

Qev, Twb
Flow rates and temperatures, etc.

Qev

Predictive model for Tw,cd,sup

Tw,cd,sup Ma,z1, Ta,in, Tchws,i, Qz1 Valve position

Chiller sequence controller

Nch Valve position Ma,z2, Ta,in, Tchws,i, Qz2 Mw,z2,bench, Tchws,bench

Zone 1

PD set-point optimizer
Mw,z1,ahx,bench, PDset,z1,bench Tchws,bench After HX Tchws,i,Qev,i & Tw,cd,sup

PD set-point optimizer
PDset,z2,bench

Zone 2

Global AHU model

Mw,z1,ahx,i

Global AHU model

Mw,z2,i

HX sequence controller
Nhx,i HX model Mw,z1,bhx,i Before HX

PD set-point adjustor
PDset,z1,adj,i

Chiller model

PD set-point adjustor
PDset,z2,adj,i

Pump sequence controller

Npu,z1,ahx,i

Wch,i

Pump sequence controller

Npu,z2,i

Water network & pump models

Wpu,z1,bhx,i

Water network & pump models

Wpu,z1,ahx,i

Water network & pump models

Performance prediction

Wpu,z2,i

Cost estimation & optimization algorithm

(Tchws,Nhx,Nch,Npu,z1,ahx,Npu,z2,PDset,z1,PDset,z2,Wtot)

Optimization process

Supervisory control strategy

Nch

Npu,z1,ahx

Npu,z2

Nhx

PDset,z1

PDset,z2

Tchws

Interface

Chiller plant control system (BAS)

Fig. 2. Illustration of optimization procedures of the optimal control strategy.

3.2. Denitions of the cost function The systematic optimization of the chilled water system aims at minimizing the total power consumption of chillers and chilled water pumps while fullling the cooling demands of all terminal units. When the evaporating temperature rises as a result of increased chilled water supply temperature set-point, the power consumption of chillers will be reduced as the suction pressure rises. However, as the chilled water supply temperature set-point increases, the heat transfer efciency of terminal units will be deteriorated. Therefore, more chilled water will be required to compensate the efciency deterioration, which will increase the power consumption of pumps. In contrast, a lower chilled water supply temperature set-point can save the power consumption of pumps, but it deteriorates the efciency of chillers, which results in more power consumption of chillers. Therefore, the chilled water supply temperature set-point should be optimized and the optimal control strategy for the overall chilled water system is to seek the most energy efcient control settings that minimize the power consumption of the entire chilled water system. Therefore, the cost

function for the system under investigation (including Zones 1 and 2) can be expressed mathematically as in Eq. (1).

J minTchws Wtot

Nch X k1

Wch;k

N1 X m1

Wpu;z1;bhx;m
N3 X j1

N2 X i1

Wpu;z1;ahx;i

Wpu;z2;j

Where J is the cost function, W is the power consumption, T is the temperature, N is the number, and subscripts ch, chws, tot, pu, z1, z2, bhx and ahx represent chiller, chilled water supply, total, pump, Zone 1, Zone 2, before heat exchangers (the primary side of heat exchangers) and after heat exchangers (the secondary side of heat exchangers), respectively. 3.3. Denition of the operating constraints The operation of the chilled water system has to obey a number of constraints, i.e., basic energy and mass balances, mechanical

1192

Z. Ma, S. Wang / Building and Environment 44 (2009) 11881198

limitations, etc. The heat generated in chiller condensers is assumed to be equal to the sum of the evaporator cooling energy and the power consumed by chillers as shown in Eq. (2). The chilled water is assumed to be distributed to each operating chiller evenly to allow all chillers to operate at good efciency. It is also assumed that the chilled water is distributed to each operating pump or each operating heat exchanger with parallel installations evenly to achieve good efciency of pumps or heat exchangers. To avoid the low chilled water supply temperature set-point causing the problems of ice in evaporators and the low efciency of chillers, and the high chilled water supply temperature set-point resulting in the problems of the humidity control for the air-conditioned spaces and inadequate cooling load provided, the chilled water supply temperature set-point is constrained between 5  C and 10  C. The lower limit of the condenser water supply temperature set-point is bounded to 18  C to avoid the low temperature set-point causing the low pressure problems in chillers. The input frequency to variable speed pumps is bounded between 20 Hz and 50 Hz. The pressure differential set-point for Zone 1 is bounded between 80 kPa and 215 kPa, while the pressure differential set-point for Zone 2 is bounded between 90 kPa and 230 kPa considering their design cooling loads and pump heads at the design condition.
Nch X k1

their service life in the long run. In addition, the minimal water ow rates in the primary and secondary sides of heat exchangers are inherently constrained by the low bound of the pump operating frequency (e.g., 20 Hz). Taking into account all these factors, a simple sequence strategy was designed and used to control the operation of heat exchangers in Zone 1. Using this strategy, another heat exchanger is activated when the water ow rate of each operating heat exchanger exceeds 80% of its design value. One of the operating heat exchangers is deactivated if the water ow rate can be handled by the remaining heat exchangers at 80% or below 80% of their design values. 3.4.4. Speed control strategy for pumps distributing water to terminals units In this study, the speeds of pumps distributing water to terminal units were controlled through resetting the pressure differential set-point at the critical loops. Fig. 3 illustrates the detailed control logic. In this strategy, the pump pressure differential set-point is optimized based on the online opening signals (i.e., PID outputs) of all water control valves. The set-point is set enough and just enough for the most heavily loaded loops. At this situation, one of control valves keeps almost fully open at all times. For reverse-return systems, a pressure differential sensor is installed at each end of the loop, as shown in Fig. 3. A representative measured pressure differential (PD), which is actually used to perform the pump speed control, is selected based on the larger deviation of measurements of both pressure differential sensors from the set-point optimized [3]. 3.4.5. Speed control strategy for pumps distributing water to heat exchangers For variable speed pumps distributing water to heat exchangers (i.e., the pumps at the primary side of heat exchangers), their speeds can be controlled using a cascade controller, as illustrated in Fig. 4. In this controller, a water ow meter is installed on the primary side of heat exchangers and a temperature sensor is installed on the outlet pipeline of the secondary side of heat exchangers. The temperature monitored by the temperature sensor is used to determine the required water ow rate with respect to the temperature set-point and the required water ow rate is then compared with the measured water ow rate in the primary side of heat exchangers to carry out the pump speed control. In this controller, a predetermined function is embedded to set the setpoints of the outlet water temperature at the secondary side of heat exchangers and the inlet water temperature at the primary side of heat exchangers with proper differences, allowing the water ow

Qcd;k

Nch X k1

Qev;k

Nch X k1

Wch;k

(2)

Where subscripts ev and cd represent evaporator and condenser, respectively. 3.4. Local control strategies The local control strategies used to design this optimal control strategy consist of the sequence control strategies for chillers, heat exchangers and pumps, and the speed control strategies for pumps with different installations. 3.4.1. Sequence control strategy for chillers The sequence control strategy for chillers is to control the number of chillers in operation. In this study, a conventional sequence strategy was used. Using this strategy, an additional chiller is brought online when the operating chillers have insufcient capacities to meet the current load. One of the operating chillers is brought ofine when the current load can be met with the remaining chillers. 3.4.2. Sequence control strategy for variable speed pumps The pump sequence control strategy is to control the number of pumps operating. In this study, a simple strategy was used as follows: bring another pump online when the frequencies of the operating pumps exceed 40 Hz. One of the operating pumps is switched off if the remaining pumps can handle the system water ow rate and head requirements at the frequency of 40 Hz or below 40 Hz. 3.4.3. Sequence control strategy for heat exchangers In this complex chilled water system, the heat exchanges are used in the water distribution systems. The operation of these heat exchangers has signicant impacts on their service life and on the power consumptions of both pumps in the primary and secondary sides of heat exchangers. Therefore, the heat exchangers should be sequenced properly. In general, operating more heat exchangers can increase their heat transfer areas and reduce the power consumptions of both pumps in the primary and secondary sides of heat exchangers to some extent due to the reduction of the total pressure drop on the heat exchangers. However, operating more heat exchangers will increase their maintenance costs and shorten

Representative measured PD Pressure controller PD set Point Pressure differential set-point optimizer

CHILLER

CHILLER

AHU

AHU

PD

AUH PID

PID

Fig. 3. The speed control strategy for variable speed pumps distributing water to terminal units.

PID

PD

Z. Ma, S. Wang / Building and Environment 44 (2009) 11881198

1193

Secondary side of HX
From terminal units HX To terminal units

Primary side of HX

0:8 C4 Mw;cd C5 Qev Wcom 1=3 C6

1 UAcd

(4) (5) (6) (7)

HX

From primary system To primary system

Mref;fic Wfic

Qev h0 h0 1 3 0 Mref;fic , h2 h0 1

Temperature Temperature set-point controller

Water flow set-point

2 Wcom a0 a1 Wfic a2 Wfic

Water flow controller

Fig. 4. The speed control strategy for variable speed pumps distributing water to heat exchangers.

Where M is the mass ow rate, UA is the overall heat transfer coefcient, h is the enthalpy, C1C6 and a0a2 are coefcients, and subscripts w, com, ref and c represent water, compressor, refrigerant and ctitious, respectively. 3.5.2. Global AHU model To predict the required chilled water ow rates of each individual zone under different operating conditions, a ctitious global AHU model is assumed and used to represent all terminal units in the zone of concern. Since the required water ow rate of the entire zone for a given condition is strongly dependent on the cooling load, zone air ow rate and AHU inlet air and water temperatures, an empirical formula as shown in Eq. (8) is used in this study to predict the performance variations of the ctitious global AHU associated with the changes of major variables. Fig. 6 presents the performance validation results of this model using the operation data collected from a simulated virtual environment representing the complex central chilling system in the super high-rise building under study. Acceptable performance predictions can be found between the model predicted water ow rates and measured values.

rates at both sides to remain roughly the same as needed for the proper control of the system. 3.5. Component models used in online optimization To achieve the desirable and satisfactory control performance for real-time applications, a number of simplied models were used in this study to formulate the optimal control strategy. A brief summary of these models is presented as follows. 3.5.1. Chiller model The chiller model utilized is developed based on the fundamental principles of thermodynamic and the heat/mass transfer processes in the chillers [12]. A ctitious refrigeration cycle (10 -20 -30 -40 ), as shown in Fig. 5, is assumed to simplify the complicated thermodynamic processes occurring in the refrigeration system. The chiller evaporator and condenser are simulated using the classical heat exchanger efciency method. The overall heat transfer coefcients of the evaporator and condenser are represented empirically as Eqs. (3) and (4), respectively. The ctitious refrigerant mass ow rate is calculated using Eq. (5). The ctitious power, as in Eq. (6), is the product of the ctitious refrigerant mass ow rate and the enthalpy difference between point 20 and point 10 in the ctitious refrigeration cycle. A polynomial as in Eq. (7) is used to characterize the relationship between the actual chiller power consumption and the ctitious power consumption.
0:8 0:745 C1 Mw;ev C2 Qev C3

d Mw d0 ,Q d1 ,Ma d2 , Ta;in Tw;in 3

(8)

Where d0d3 are coefcients, and subscripts a and in represent air and inlet, respectively. 3.5.3. Heat exchanger and pump models The performance of the water-to-water heat exchanger is modeled using 3-NTU method. The actual heat transfer in the heat exchanger is computed using Eq. (9). The heat transfer effectiveness of the counter ow heat exchanger is simulated using Eq. (10). The overall number of transfer units (NTU) is dened as in Eq. (11). The overall heat transfer coefcient (UA) is simulated using Eq. (12) taking into account the inuences of the variations of water ow rates in both sides of heat exchangers.

1 UAev

(3)

Q 3,Cmin , Th;w;in Tc;w;in

(9) (10)

Pressure Qcd = Qev + Pcomp Pcd 3 3 2 2

1 expNTU1 u 1 uexpNTU1 u

600

Model predicted water flow rate (l/s)

500 400 300 200 100 0

Pcomp Pev 4 4 Qev Enthalpy

Fig. 5. Schematic of the pressureenthalpy diagram (actual cycle: 1-2-3-4; ctitious cycle: 10 -20 -30 -40 ).

100

200

300

400

500

600

"Measured" water flow rate (l/s)

Fig. 6. Performance validation results of the ctitious global AHU model.

1194

Z. Ma, S. Wang / Building and Environment 44 (2009) 11881198

NTU

UA Cmin !m Mh , Mh;des !n

(11)

Mc UA UAdes , Mc;des

(12)

Where 3 is the heat transfer effectiveness, C is the capacity ow rate, u is the ratio of the minimum capacity ow rate to the maximum capacity ow rate, m and n are indexes, and subscripts h, c and des represent hot, cold and design, respectively. The pump model used was comprised of polynomials representing head versus ow and speed, efciency versus ow and speed, motor efciency versus the fraction of the nameplate brake horsepower, and VFD (variable frequency driver) efciency versus the fraction of the nominal speed. The head and efciency characteristics are based on manufacturers data at the full speed operation and extended to variable speed operation using pump afnity laws [13]. 3.5.4. Water network pressure drop model To design the optimal control strategy, a water network pressure drop model that characterizes the pressure drop on each individual component in the system of concern is developed. For the water subsystems with different congurations, the water network pressure drop models will be signicantly different. Since the water subsystems of the secondary side of heat exchangers with pumps distributing water to terminal units could be the most complicated systems, the other water subsystems can be considered as the simplications of such systems. Therefore, a water network pressure drop model for such systems is presented in the following briey. Fig. 7 illustrates the simplied structure of the water network pressure drop model for the subsystem at the secondary side of heat exchangers in Zone 1, in which only 6 terminal units are illustrated for example. The overall pressure drop of this system, i.e., along the sub-branch FF1, can be mathematically described as in Eq. (13), which includes the pressure drop on heat exchangers (including the pressure drops on heat exchangers and on the headers that direct the ow into and from each heat exchanger), the pressure drop on the ttings around pumps (including the pressure drop on the headers that direct the ow into and from each pump and the pressure drop on the valves in pump headers), the pressure drop on main supply and return pipelines, the pressure drop across the sub-branch (i.e., FF1) and the pressure drops on the pipeline sections of AB, BC, CD, DE and EF.

Eq. (13) can be further re-written as in Eq. (14) by representing the water ow rate in each pipeline section (i.e., AB, BC, CD, DE and EF) as a function of the total system water ow rate by multiplying a converting factor (i.e., f1f5). For simplifying the calculations, the factors (f1f5) in the same zone can be considered constant under various load conditions since the usage characteristic and load prole of each oor in the same zone are similar in the building under study. The overall system pressure drop can therefore be nally expressed as in Eq. (15).

PD

Shx 2 Spf 2 2 2 2 M 2 M S0 S11 M0 S1 M1 S2 M2 2 Nhx 0 Npu 0

2 2 2 S3 M3 S4 M4 S5 M5 PDFF1

13

PD

 Shx 2 Spf 2 2 M0 2 M0 S0 S11 M0 S1 f2 S2 f2 1 2 2 Npu Nhx  2 S3 f2 S4 f2 S5 f2 M0 PDFF1 3 4 5 o Shx 2 Spf 2 n 2 2 M0 2 M0 S0 S11 M0 Sfic M0 PDFF1 2 Npu Nhx

14

PD

(15)

Where PD is the pressure drop, S is the ow resistance, and subscripts hx and pf represent heat exchanger and pump ttings, respectively. The last term on the right-hand side of Eq. (15) is the optimal pressure differential set-point (named PDset) at a given chilled water supply temperature set-point. The other three terms are the pressure drops across the heat exchangers, the ttings around pumps and the distribution pipelines, respectively. These three pressure drops are named piping pressure drop (PDpiping). The overall system pressure drop can therefore be divided into two parts: PDset and PDpiping, as shown in Fig. 8. It can be observed that the pump operation point is the intersection of both the pump curve and control curve. It is worthy noticing that both PDset and PDpiping in Fig. 8 vary with the changes of working conditions. The piping head loss curve is not constant, which varies with the changes of the number of pumps in operation and/or the number of heat exchangers in operation. For online applications, there are ve parameters (Shx, Spf, S0, S11 and Sc) required to be identied. In fact, the parameters of S0, S11 and Sc can be integrated as a single parameter (Sc0 ) to represent the ow resistance on the distribution pipelines, including the main supply and return pipelines and the pipeline sections of AB,

Spf,tot =

Spf Npu2
S0 S1 M1 SA S2 M2 SB S3 M3 SC S4 M4 SD S5 M5 SE SF

AHU AHU

AHU

AHU

AHU

AHU

AHU

PD1

Shx,tot =

Shx Nhx2

HX

HX

S6 M0 S11

S7

S8

S9

S10

A1

B1

C1

D1

E1

F1

Fig. 7. Structure of the water network pressure drop model.

PD2

Z. Ma, S. Wang / Building and Environment 44 (2009) 11881198

1195

PD set-point ( kPa)

Head

PDset

PDset,max PDset PDset,bench

PDset

Pump curves

Mw

PDset,min

Pump head

Pip

ing

he

oss ad l

cur

PDpiping

t Con

ro

rv l cu

e
ve

Mw,low

Mw,bench

Mw

Mw,up
System flow (L/s)

Fig. 9. Working principle of the incremental model.

Flow rate
Fig. 8. Schematics of the system control curve and piping head loss curve as well as pump curves.

properly, a low limit and an upper limit are constrained on the pressure differential set-point, as shown in Fig. 9. This model was also validated using the operation data collected from the simulated virtual environment representing the complex central chilling system in the super high-rise building under study. Fig. 10 gives the validation results. It can be observed that this model has acceptable performance in prediction. 3.5.6. Predictive model for the condenser water supply temperature set-point Since the condenser water supply temperature set-point is one of major control variables affecting the power consumption of chillers, a predictive model, as shown in Eq. (17), was used in this strategy to predict a near optimal condenser water supply temperature set-point for a given condition. The coefcients in this model were determined based on complete simulations. The values of these coefcients used are given as follows: e0 7.33666, e1 0.79998, e2 3.78266.

BC, CD, DE and EF. The parameters of Shx and Spf can be determined using the measured pressure drops across the heat exchanger sub-circuit and pump tting sub-circuit together with the measured system water ow rate. The parameter of Sc0 can be determined using the system water ow rate, pump head and the pressure differential set-point at the design condition as well as the identied parameters of Shx and Spf. 3.5.5. Pressure differential set-point incremental model To formulate the optimal control strategy, a pressure differential set-point incremental model was developed to predict the optimal pressure differential set-points under different chilled water supply temperature set-points for a given condition. In fact, at a given chilled water supply temperature set-point, the water ow rate required by the terminal units in the zone of concern is unique if the cooling load and inlet and outlet air status of terminal units keep unchanged. Therefore, the pressure differential set-point under any given chilled water supply temperature set-point can be characterized by using the system water ow rate, and the basic form of the incremental model can be expressed using a secondorder polynomial as shown in Eq. (16) in terms of the water ow rate instead of using the temperature set-point.

Tw;cd;sup e0 e1 Twb e2

Qev Qev;des

! (17)

Where e0e2 are coefcients, and the subscript sup indicates supply. 3.6. Optimization procedures At each given condition, this optimal control strategy seeks the optimal control settings following procedures (also illustrated in Fig. 2) as below. (1) Validate the measurements and control signals using a lter; (2) Determine a near optimal condenser water supply temperature set-point and the number of chillers operating according
150

2 PDset PDset;bench g0 Mw Mw;bench g1 Mw Mw;bench g2

16

Where g0g2 are coefcients, and subscripts set and bench indicate set-point and benchmark, respectively. Fig. 9 illustrates the basic working principle of this pressure differential set-point incremental model. For a given condition, the collected opening signals of all water control valves concerned are used to determine an optimal pressure differential set-point at the corresponding measured chilled water supply temperature setpoint. The system water ow rate at this measured chilled water supply temperature set-point can also be measured and collected. The optimal pressure differential set-point and the water ow rate at this measured chilled water supply temperature set-point are then used as the benchmarks (PDset,bench, Mw,bench), as shown in Fig. 9, to predict the pressure differential set-point at any other system water ow rate, which corresponds to a given chilled water supply temperature set-point and is predicted by the ctitious global AHU model. This process is named PD set-point adjustor in Fig. 2. To deliver the adequate chilled water to satisfy the cooling demands of all terminal units and to allow the system to operate

Predicted (PDset-PDset,bench) (kPa)

100 50 0 -50 -100 -100

-50

50

100

150

'Measured' (PDset-PDset,bench) (kPa)

Fig. 10. Validation results of the incremental model.

1196

Z. Ma, S. Wang / Building and Environment 44 (2009) 11881198

(3)

(4)

(5)

(6)

(7)

(8) (9)

(10)

to Eq. (17) and the chiller sequence control strategy (Section 3.4.1), respectively; Determine the optimal pressure differential set-points at the given chilled water supply temperature set-point for Zone 1 and Zone 2, respectively; Estimate the optimal pressure differential set-point and water ow rate at a given chilled water supply temperature setpoint; Determine the number of heat exchangers in Zone 1 in operation based on the predicted water ow rate of Zone 1 at the given chilled water supply temperature set-point; Predict the operating number of pumps in the secondary side of heat exchangers in Zone 1, the operating number of pumps in Zone 2 and their power consumptions; Determine the outlet water temperature at the secondary side of heat exchangers in Zone 1 and predict the water ow rate and the power consumption of the pump in the primary side of heat exchangers in Zone 1; Predict the power consumption of operating chillers; Determine the most energy efcient operating combination (i.e., control settings) that minimizes the overall system energy consumption after the power consumptions of all possible operating combinations have been predicted by using the cost estimator and optimization algorithm; The supervisory strategy is nally used to provide the nal control settings taking into account the operating constraints and limitations of practical applications.

12000 10000 Zone 1 Zone 2

Cooling load (kW)

8000 6000 4000 2000 0

10

12

14

16

18

20

22

24

Time (h)
Fig. 11. Cooling load proles in Zones 1 and 2 in the typical mild-summer day.

observed that the cooling load proles in both zones are signicantly different due to the different usage characteristics of both zones. 4.2. Test and evaluation results of the strategy The energy performance of this optimal control strategy (namely Strategy #4) was evaluated by comparing it with that of other reference control strategies (Strategies #1, #2 and #3) as listed in Table 2. It is worth noticing that the same performance predictor, cost function, optimization technique, supervisory strategy and local control strategies except the speed control strategy for pumps distributing water to terminal units as the proposed strategy utilized were used in the other three strategies. The xed chilled water supply temperature set-point used was the design temperature set-point of 5.5  C. The xed pressure differential set-points used for Zone 1 and Zone 2 were the upper limits of the pressure differential set-points bounded in Section 3.3. In the following, the performance of Strategy #1 was used as the benchmark for comparisons. Table 3 summarizes the daily electrical energy consumptions of pumps (except the primary chilled water pumps) and chillers as well as the total electrical energy consumption of chillers and pumps in the chilled water system of concern (only including Zone 1 and Zone 2) in the three typical days under different control strategies. Compared with Strategy #1 using xed setpoints, Strategy #4 using optimal set-points saved around 1028.7 kW h (24.86%), 989.3 kW h (19.14%) and 801.7 kW h (11.99%) of the pump energy, and 986.9 kW h (2.63%), 1012.7 kW h (2.27%) and 682.5 kW h (1.28%) of the total energy in the typical spring day, mild-summer day and sunny-summer day, respectively. It also can be found that Strategy #2 saved about 103.6 kW h (0.28%), 149.9 kW h (0.34%) and 143.3 kW h (0.27%) of the total energy, while Strategy #3 saved about 836.7 kW h

4. Performance tests and evaluation of optimal control strategy 4.1. Setup of the tests Since the building under study is still being constructed, the hourly-based annual cooling load proles of each individual zone of this building were calculated using EnergyPlus based on the design data and hourly-based weather data for the typical year in Hong Kong. To calculate the air ow rates and inlet air dry-bulb temperatures of terminal units in each individual zone, a series of assumptions were used as follows. A minimum ratio of the fresh air is assumed to be 20%. The room design air dry-bulb temperature is 23  C for summer and autumn cases, and 21  C for winter and spring cases with a 50% relative humidity. The air dry-bulb temperature leaving terminal units is controlled at 13  C with a 95% relative humidity. During the tests, the online opening signals of water control valves were collected from the simulated virtual environment representing the complex central chilled water system in the super high-rise building under study. The weather data, including the air wet-bulb temperature and air dry-bulb temperature, and cooling loads of each individual zone were provided by the data les to the simulated virtual environment as the test condition. Three typical days were selected to test and evaluate the energy performance of the proposed optimal control strategy. These days represent the typical operating conditions of the air-conditioning system in the typical spring, mild-summer and sunny-summer days, respectively. The cooling loads in Zone 1 varied from 870 kW to 6000 kW, from 1000 kW to 6500 kW and from 1120 kW to 7300 kW, while the cooling loads in Zone 2 varied from 970 kW to 8900 kW, from 1320 kW to 9800 kW and from 1540 kW to 11,500 kW in the selected typical spring day, mild-summer day and sunnysummer day, respectively. Fig. 11 presents the cooling load proles in both zones in the typical mild-summer day. It can be

Table 2 Description of the control strategies tested and evaluated. No. Strategies Descriptions 1 2 3 4 Strategy #1 Strategy #2 Strategy #3 Strategy #4 Using the xed chilled water supply temperature set-point and the xed pressure differential set-point at the critical loops Using the optimal chilled water supply temperature set-point and the xed pressure differential set-point at the critical loops Using the xed chilled water supply temperature set-point and the optimal pressure differential set-point at the critical loops Using the optimal chilled water supply temperature set-point and the optimal pressure differential set-point at the critical loops

Z. Ma, S. Wang / Building and Environment 44 (2009) 11881198 Table 3 Comparison of the daily electric energy consumptions under different control strategies. Control strategies Wpu (kW h) Wch (kW h) Wpu Wch (kW h) Wpu saving kW h Spring case Strategy #1 Strategy #2 Strategy #3 Strategy #4 Mild-summer case Strategy #1 Strategy #2 Strategy #3 Strategy #4 Sunny-summer case Strategy #1 Strategy #2 Strategy #3 Strategy #4 4138.3 4213.5 3301.6 3109.6 5169.5 5311.7 4321.4 4180.2 6687.2 6796.4 6174.0 5885.5 33,421.8 33,243.0 33,421.8 33,463.6 39,436.6 39,144.5 39,436.6 39,413.1 46,654.4 46,401.9 46,654.4 46,773.6 37,560.1 37,456.5 36,723.4 36,573.2 44,606.1 44,456.2 43,758.0 43,593.3 53,341.6 53,198.3 52,828.4 52,659.1 75.2 836.7 1028.7 142.2 848.1 989.3 109.2 513.2 801.7 % 1.82 20.22 24.86 2.75 16.41 19.14 1.63 7.67 11.99 Wch saving kW h 178.8 0.0 41.8 292.1 0.0 23.5 252.5 0.0 119.2 0.54 0.00 0.26 % 0.53 0.00 0.13 0.74 0.00 0.06 Total saving kW h 103.6 836.7 986.9 149.9 848.1 1012.8 143.3 513.2 682.5

1197

% 0.28 2.23 2.63 0.34 1.90 2.27 0.27 0.96 1.28

(2.23%), 848.1 kW h (1.90%) and 513.2 kW h (0.96%) of the total energy in these three test days, respectively, as compared with Strategy #1. From Table 3, it also can be observed that the strategies using the optimal pressure differential set-points (Strategies #3 and #4) can save signicantly more energy than the strategies using the xed pressure differential set-points (Strategies #1 and #2). Compared with Strategy #1, the savings using Strategy #3 were all contributed by using the optimal pressure differential set-points only. Fig. 12 presents the hourly-based power consumptions of the pumps using Strategies #1 and #3 in the typical mild-summer day to demonstrate more details of the energy saving potentials using the optimal pressure differential set-points in the system under study. Compared with the strategies using the xed chilled water supply temperature set-point, the energy saving potentials related to the use of the optimal temperature set-points were very limited in this particular building. Only about 0.270.34% of the total electric energy consumption in the system studied can be saved. The reasons are that the usage characteristics and load proles of Zones 1 and 2 are different (see Fig. 11), and the inlet water temperatures of terminal units in Zones 1 and 2 are signicantly different due to the application of the heat exchangers in Zone 1. Therefore, the actual optimal chilled water supply temperature setpoints for Zone 1 and Zone 2 could be signicantly different. The optimal chilled water supply temperature set-points identied for

the overall system were the tradeoffs of the power consumptions among chillers, pumps in Zone 1 and pumps in Zone 2. Fig. 13 presents the hourly-based total power consumptions of the chilled water system under investigation by using Strategy #1 (the xed temperature set-point) and Strategy #2 (the optimal temperature set-point) in the typical mild-summer day. The power savings of using Strategy #2 over Strategy #1 are also illustrated in this gure. It can be seen that the power consumption proles using both strategies were very similar since the optimal chilled water temperature set-points found by Strategy #2 were very close to the xed temperature set-point of 5.5  C for most cases. Based on the above comparisons, it can be concluded that the proposed optimal control strategy using both optimal temperature set-points and optimal pressure differential set-points can save signicantly more energy than the other three control strategies. It is also demonstrated that most of energy savings in the proposed strategy were contributed by using the optimal pressure differential set-points. The energy saving potential associated with the use of the optimal chilled water temperature set-points in the building under study was very limited and, therefore, the resulted total energy saving percentage due to the application of this optimal strategy in this particular building is relatively small. However, this optimal control strategy can be easily extended to most of the chilled water systems with primarysecondary pumping paradigms since most issues associated with the control of chilled water systems are included in this strategy.

400 350

Pump power (kW)

300 250 200 150 100 50 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24

Total power consumption (kW)

Strategy #1 Strategy #3

3200 2800 2400 2000 1600 1200 800 400 0 0 2 4 6 8 10 12 14 16 18 20 22 Strategy #1 Strategy #2 Power saving

25 20 15 10 5 0 24

Power saving (kW)

Time (h)
Fig. 12. Comparison of the pump powers between the strategies using the xed and optimal pressure differential set-points in the typical mild-summer day.

Time (h)
Fig. 13. Comparison of the total power consumptions between the strategies using the xed and optimal temperature set-points in the typical mild-summer day.

1198

Z. Ma, S. Wang / Building and Environment 44 (2009) 11881198

5. Conclusions An optimal control strategy for complex building central chilled water systems is developed for real-time applications. This optimal control strategy consists of the model-based performance predictor, cost estimator, optimization technique, supervisory strategy and a number of local control strategies. The performance of this strategy was evaluated on the complex central chilled water system in a super high-rise building in Hong Kong by simulation tests. Results show that energy in the system under study can be saved associated with the use of this optimal strategy as compared with the other three conventional or xed set-point control strategies. It is observed that around 11.9924.86% of the pump energy and 1.282.63% of the total electric energy in the system under investigation can be saved, respectively, by using this optimal strategy as compared with the strategy using the xed temperature and pressure differential set-points. Although the percentage of saving in terms of total energy is relatively small when recognizing the modeling mismatch, and the performance benets in practice might likely be hard to measure, however, it can be expected that the actual energy saving using this proposed strategy in practice could be more than the energy saving presented above since Strategy #1 using the xed set-points used as the baselines in this study has already somewhat optimized. This is because Strategy #1 used the same cascade controller as that of the other three strategies used to control the pump in the primary side of heat exchangers and the pump speeds distributing water to terminal units were controlled using the pressure differential at the critical loops instead of using the pressure differential at the main supply and return pipelines. This optimal control strategy is still simple and easy to implement in practice as well. These characteristics make it suitable for online control applications. This strategy is being implemented in the super high-rise building for eld application and validation.

Acknowledgements The research work presented in this paper is nancially supported by a grant (PolyU5308/08E) of the Research Grant Council (RGC) of the Hong Kong SAR and the support of Sun Hung Kai Real Properties Limited. The authors would like to thank Mr. W.K. Pau, Sun Hung Kai Real Properties Limited for his support to the research work.

References
[1] Rishel JB. Control of variable-speed pumps on hot-and chilled-water systems. ASHRAE Transactions 1991;97(1):74650. [2] Rishel JB. Water pumps and pumping systems. New York: McGraw-Hill; 2002. [3] Rishel JB. Control of variable speed pumps for HVAC water systems. ASHRAE Transactions 2003;109(1):3809. [4] Moore BJ, Fisher DS. Pump pressure differential setpoint reset based on chilled water valve position. ASHRAE Transactions 2003;109(1):3739. [5] Ahmed O. DDC applications in variable-water-volume systems. ASHRAE Transactions 1991;97(1):7518. [6] Braun JE, Klein SA, Beckman WA, Mitchell JW. Methodologies for optimal control to chilled water systems without storage. ASHRAE Transactions 1989;95(1):65262. [7] Braun JE, Klein SA, Mitchell JW, Beckman WA. Applications of optimal control to chilled water systems without storage. ASHRAE Transactions 1989;95(1):66375. [8] Cascia MA. Implementation of a near-optimal global set-point control method in a DDC controller. ASHRAE Transactions 2000;106(1):24963. [9] Lu L, Cai WJ, Xie LH, Li SJ, Soh YC. HVAC system optimization in-building section. Energy and Buildings 2005;37(1):1122. [10] Jin XQ, Du ZM, Xiao XK. Energy evaluation of optimal control strategies for central VWV chiller systems. Applied Thermal Engineering 2007;27(56):93441. [11] Wang SW, Ma ZJ. Supervisory and optimal control of building HVAC systems: a review. HVAC&R Research 2008;14(1):332. [12] Ma ZJ, Wang SW, Xu XH, Xiao F. A supervisory control strategy for building cooling water systems for practical and real time applications. Energy Conversion and Management 2008;49(8):232436. [13] Bahneth WP, Peyer E. Comparative analysis of variable and constant primaryow chilled-water-plant performance. Heating, Piping, Air Conditioning Engineering 2001;73(4):4150.