Pergamon PII: S03604323(97)0001~X

&i/ding ondEmironmenr, 1997 Published by Elsewer

Vol. 33, No. 1, pp. I-20, 1998 Science Ltd. All rights reserved Printed in Great Britain 0360-1323198 %19.OO+O.Oil

Dynamic Simulation of a Building Central Chilling System and Evaluation of EMCS On-Line Control Strategies
SHENGWEI WANG*
(Received 20 January 1997; accepted 30 April 1997)

Dynamic models of centrifugal chillers, heat exchangers, seawater and chilled-water networks, cooling coil, actuator, sensor, variable-speed pump and DDC controller of EMCS are developed to simulate the dynamics of a seawater-cooled chilling system controlled by EMCS on-line strategies. The thermal, hydraulic, energy and control performances of the system are simulated. The on-line control strategies (i.e. adaptive and derivative strategies) developedfor the central chilling system are tested and evaluated by applying them to control the simulated living chilling system under different AHU dynamic loads. This paper presents the models, system dynamic simulation of the chilling system, chilling system performance monitoring, validation of simulation, EMCS on-line control strategies and evaluation of the strategies. 0 1991 Published by Elsevier Science Ltd.

NOMENCLATURE
impeller exit area, area [m] velocity [m/s] CP specific heat capacity [kJ/(kg K)] capacity flow rate, thermal capacity [kW] C frequency [Hz] enthalpy [kJ/(kg K)] i hydrodynamic losses [kW/kg] hW compression work [kW/kg] h PO] polytropical compressor theoretical head [kW/kg] h mass flow rate [kg/s] ; number of transfer units [-_I NTU pressure [kPa] P valve position [-_I P heat transfer rate [kW] Q R heat transfer resistance [k/kW] sensible heat ratio [-_I SHR temperature [Cl t, T impeller tip peripheral speed [m/s] UA overall heat transfer coefficient [w/K] specific volume [m3/kg] ct: power consumption [kW] Greek letters impeller vane angle B 6 heat exchanger efficiency 0 prerotation vane angle t time [s] Subscripts a air C coil cd condenser or condensation camp compressor ev evaporator or evaporation A c

ex

exch f imp in max min pi Pu sea tot

W wt

exhaust heat exchanger fictitious flow impeller inlet maximum minimum pipe pump seawater total water wet regime

INTRODUCTION Dynamic been simulation of a building HVAC system has

*Department of Building Kong Polytechnic University,

Services Engineering, The Hong Kowloon, Hong Kong. 1

used in assisting the building control application in three areas as summarised by Kelly and Dexter in a comprehensive literature review [ 1, 21: regulation, supervisory control and optimised building control. In practice, these topics will tend to overlap one another. When simulation aims at studying or optimising the regulation of HVAC processes (local loop control strategies), only the dynamics of certain processes or subsystems is of concern. The studies on control performance and adaptive controls of air handling unit (AHU) and variable air volume (VAV) are examples of such an application [3-91. The IEA research project Annex 17 gives an example reflecting the advance of system simulation in testing and evaluating energy management and control systems (EMCS) supervisory control strategies for entire building systems as well as control strategies implemented in real

Shengwei Wang and supervisory strategies in different seasons. The energy and control performances of derivative and adaptive control strategies for the on-line seawater pressure set-point reset were tested and evaluated.

EMCS by means of emulation [IO]. The application can be further divided into two categories: energy per.fbrmunce of control strategies, and realistic control perfbnance of control strategies. When only the energy performance of the strategies is of concern, the dynamics of processes, such as building thermal dynamics and ice storage, which noticeably affect the energy performance of the system, are modelled by simulation [I I-131. The dynamics of the components such as sensors, actuators, and coils only affects the control performance but do not affect the energy performance. The dynamics of those components is not modelled in the simulation. When the realistic performance of control strategies is of concern, the dynamics of the system including not only the building thermal dynamics but also the sensors, coil, etc. which affects the local loop control activities should be properly modelled in the simulation [l&18]. In Annex 17, however, the hydraulic loop pressure flow balance was neglected and dynamics of the hydraulic loop was idealised since the pressure control of air and water loops was not of concern in the studies. In fact, most well-known HVAC system dynamic simulation programs, such as TRNSYS, HVACSIM+, BLAST, do not provide sufficient component models to simulate the dynamic balance of the hydraulic network of a HVAC system, particularly for large water or air networks. The programs (e.g. TRNSYS) might have difficulty in reaching convergence in large hydraulic network pressure flow balance computations when a component model concept similar to the thermal models is applied in modelling the networks, due to the sensitivity of the hydraulic balance. Although there are tools simulating the dynamics of large water and air flow networks of the HVAC system, these tools might not consider the pressure flow dynamics of the system and other dynamic performances (i.e. thermal dynamic performance, energy performance, etc.) as a whole to simulate the performance of EMCS control strategies in real life. The study presented in this paper simulates the energy and thermal dynamic performances of the chilling system as well as the pressure flow dynamic performance of the condenser cooling seawater network and chilled-water network to provide a realistic and living artificial environment for EMCS strategy testing. The dynamic models of centrifugal chiller, heat exchanger, seawater and chilledwater networks, cooling coil, actuator, sensor, variablespeed pump and frequency inverter, DDC controllers of EMCS are developed and used to simulate the realistic (i.e. thermal dynamic, energy and control) performances of the chilling system controlled by on-line control strategies. In order to avoid the convergence difficulties in water network pressure flow balance computation, separate models are developed to simulate the dynamics balance of the chilled-water and seawater networks, respectively. The central chilling system of an existing office building is modelled and simulated. To represent the dynamic cooling load of the chilling system from the air handling units in the building, the AHUs controlled by DDC are included in the simulation. Simulation exercises were performed on the dynamic system simulated to test and evaluate the on-line control performance of EMCS local

CHILLING

SYSTEM

The central chilling system of a large office building in Hong Kong is simulated. The building has 46 storeys and a usable area of about 74,OOOm. The central chilling system is located in the basement with the seawater pump station at the sea front about 500 m away. The central chilling plant (Fig. 1) consists of five identical indirect seawater-cooled centrifugal chillers using R12, three variable-speed seawater pumps, and two auxiliary heat exchangers. Four of the chillers serve for normal duty and one for standby. The design cooling capacity of each chiller is 3100 kW. Two of the pumps are duty pumps and one is for standby. One auxiliary heat exchanger is used (the second is for standby) 24 h a day serving the essential A/C units in a number of important rooms, such as computer centres. Each chiller is associated with one constant condenser water pump and one constant primary chilled-water pump. There are four constant secondary chilled-water pumps (three for duty and one for standby). The seawater pump speed is controlled by moderating the frequency input using one frequency inverter for each of the pumps. A few office floors are in use 24 h a day, including public holidays. Therefore, the chilling system operates continuously and the load on the chillers varies significantly even within one day. The pressure control of the seawater system, the AHU supply air temperature control and the EMCS monitoring points essential for the EMCS control strategies are shown in Fig. 1. The cooling of the offices is served by the AHUs at each office floor (only the control loop of one AHU cooling coil is shown in Fig. 1). The supply air temperature after a cooling coil is controlled by a PID temperature controller in a typical AHU, which adjusts the chilled-water flow rate through the cooling coil by moderating the two-way moderating valve. The differential seawater pressure between supply and return pipes near heat exchangers is monitored and controlled by a DDC controller. The PID pressure controller controls the pressure by moderating the frequency of the AC power input to the pumps. The AC power input frequency to the two pumps was set to be the same to share the load equally. The instantaneous power consumption of each chiller and seawater pump is measured by a kilowatt transducer. The differential pressure of the supply water and return water pipes near the exchangers is measured by differential pressure sensors. The supply seawater temperature at the supply pipe is also measured. The return seawater temperature after each exchanger, the condenser cooling water temperatures at the inlet and outlet of each exchanger are measured to monitor the performance of the heat exchangers. The chilled-water temperature, the chilled-water temperatures to/from the building, the chilled-water flow rate to the building, the water flow rate through the bypass pipe, the differential pressure

Dynamic

simulation

of a building central chilling system i

7 From Building W

Chiller

Power

Transducer Sensor Pressure Sensor

Temperature Differential Flow Meter

Fig. 1. Central chilling system and chilled-water network with measurement points essential for EMCS
controls.

across the secondary pumps, rate are monitored by EMCS.

and

the seawater

flow

CHILLING

SYSTEM

COMPONENT

MODELS

Centri@gal chiller The chiller model simulates the dynamic performance of single or multi-stage water-cooled centrifugal chillers using a steady-state approach and a dynamic approach. The model simulates the chiller performance under various working conditions on the basis of the impeller tip speed (u,), impeller exhaust area (A), impeller blade angle (8) and 13 other coefficients/constants, which might be available from chiller technical data or identified (partlally or fully) by an associated preprocessor using chiller performance data under full load and partial load. The chiller chilling capacity is assumed to be controlled by adjusting the inlet vane angle (0). Figure 2a shows the refrigeration cycle of the two-stage centrifugal chiller model. Compressor. The compressor is modelled on the basis of mass conservation, Euler turbomachine equation and energy balance equation. Figure 2b shows the velocity triangles of the impeller and prerotation vane. The Euler equation is modified by considering the impeller exit radial velocity (c,J distribution, and derived as equation (1):

where h,, is the theoretical head, B is the ratio of impeller channel depth at intake to that at exhaust, D, and v, are specific volume at impeller intake and exhaust, respectively. Energy balance equations are applied to two control volumes, i.e. compressor control volume (from compressor suction to compressor exhaust) and impeller control volume (from compressor suction to impeller exit) as shown in equations (2) and (3):

h,, = h,,,

,mp

-t

h,,

nnp +

2: /

(3)

where h,,, is polytropic compression work, hhyd is hydrodynamic losses, ci is vapour velocity at impeller exhaust. Hydrodynamic losses in two control volumes are considered to be composed of three elements, i.e. flow friction losses, inlet losses and incidence losses as shown in equations (4) and (5). The flow friction losses are considered proportional to the compressor volume flow rate squared and thus proportional to the impeller exit radial velocity squared (&). The inlet losses are considered proportional to the velocity through prerotation vanes chan-

Shengwei Wang

Pcd P,

(b)

impeller

--_

-_~-

_ \\L-_L ; \
Fig. 2. Schematic
pre-rotation vane /

+-+=-=j

Cd I

of refrigeration

cycle and velocity triangles.

nel squared. Incidence losses are considered to the shock velocity component squared.

proportional

stage is assumed to have the same flow efficiency and compression ratio as the first stage. Mass and energy conservations are applied to the economiser and the mixing process at the second stage suction. Only the first stage impeller geometric parameters are of concern. Condenser and evaporator. The evaporator and condenser are simulated using the classical heat exchanger efficiency method. By considering the effects of water flow rate (MW,c,, Mw,J and heat flux (Qe,, Qcd), the evaporator and condenser overall heat transfer coefficients (UA,,, (/A_,) are represented empirically as shown by equations (6) and (7): CIA,, = [C,M,~,8+C2Qevo-74s+C3]~, UA,* = [C&f,::+ where C,-C, are constants. C,Qz+ CJ, (6) (7)

where [, $,, tj2 and x are the introduced constants. Given the evaporator pressure, condenser pressure and position of inlet vanes (value of 8), the compressor model can calculate radial velocity and specific volume at impeller exhaust, and thus refrigerant mass flow rate and internal power. The compressor capacity is controlled by the inlet vanes angle (0) as shown in equations (4) and (5). For multi-stage chillers, single stage compressor equations are used to calculate the first stage. The second

Dynamic

simulation

of a building central chilling system

The evaporation and condensation temperatures, and thus the evaporator and condenser pressures, are calculated, given the chilling capacity (Q,), heat rejection (Qcd), chilled and cooling water flow rates and inlet temperatures. Power consumption. Chiller power consumption (IV) is calculated on the basis of the compressor internal power (IV,,). With the consideration of compressor mechanical and leakage losses, motor electrical and mechanical losses, the chiller power consumption is the sum of three components, i.e. the internal compression power (W,,), a variable part of the losses which is proportional to the internal compression power, and a constant part of the losses (IV,) as shown in equation (8): W= aW,,+ W,, where a is a coefficient. Model parameter identification. Of the model parameters, impeller geometric parameters (zL~,A, /I) are either given by the manufacturer or, together with the other parameters, identified according to the performance data given by manufacturers or from field measurements, Parameter identification is carried out by a preprocessor developed for the model. Dynamics model of chiller. The dynamics of the chiller is simulated by assuming two thermal storages, one at the cooling water inlet of the condenser and the other at the chilled-water inlet of the evaporator, as shown in Fig. 3 and represented mathematically by two first-order differential equations (equations (9) and (10)). This simple approach presents both the dynamic response of the chiller to the change of working conditions (inlet temperatures) and the dynamic effects of the working condition changes on the compressor load. (9)

water at the evaporator and cooling water at the of condenser, T:,,,, and T:d,,n are the inlet temperatures chilled-water and condenser cooling water after introducing dynamic effects of evaporator and condenser. Variable-speed pump and seawater network The variable-speed seawater pump set is simulated by a steady-state pump, a steady-state frequency inverter and a dynamic actuator of the inverter. The frequency at the outlet of the inverter is linear to the input signal from the actuator. The efficiency of the inverter is included within the model of pump energy performance. The energy performance and pump characteristics at various speeds are simulated using fourth-order polynomial functions as shown in equations (11) and (12). The coefficients of the equations are determined by regression using the performance data from experiments or manufacturers catalogues.

(8)

Wpu(.L Mpu)=

f-O,=0

z i GJMpu,

(11) (12)

(IO) are the inlet temperatures whereT,,,,, and Tcd,,n of chilled-

Condenser

where WpUand PpUare the pump power consumption and pressure head, respectively,fis the frequency input to the pump, M,, is the water flow rate through a pump, and G and E are coefficients. To avoid the difficulty of convergence, which is faced in most cases when solving the equations in separate models using component-based simulation programs, the pressure flow characteristics of the entire seawater network are simulated in one single model (Fig. 4). The pressure flow balance of the seawater network is obtained by solving the pressure flow equations of the entire network within one single model by internal iteration. The thermal dynamic performance of heat exchangers is simulated by another separate model which uses the water flow rates through heat exchangers available from the outputs of the seawater network model. The model of the seawater network pressure flow characteristics consists of the pressure balance equation (equation (13)) flow balance equation (equation (14)) and the equations presenting the pressure flow characteristics of exchangers, pipe and variable-speed seawater pumps. The pressure flow characteristics of the pipes and heat exchangers are modelled by polynomial equations and the coefficients of the equations are determined using the pressure flow performance data measured on site. These coefficients might be determined by the pipe and valve pressure loss equations when details of the geographic data of the network are available. P&f; M,,) - Pp,(M,,,)Pexch,XKxch.r) = 0, (13)

Expansion Device

Compress0 #

Chilled Water O,k

Evaporator
of chiller dynamic

Chilled water Inlet I

Fig, 3. Schematic

model.

Secondary chilled-water network The pressure flow balance of the secondary chilledwater network is shown in Fig. 5. The building is divided into five zones. The pressure flow characteristics of the selected AHU from each zone are simulated and all the

Shengwei Wang

Water Pipe Fig. 4. Schematic of seawater network model.

Fig. 5. Schematic

of secondary

chilled-water

network

model.

AHUs in the same zone are considered by multiplying the flow rate of the simulated AHU by a ratio. The entire secondary chilled-water network flow pressure balance is simulated and solved by one single model, similarly to avoid difficulties in convergence. As inputs of the network model, the positions of the two-way moderating valves are given by the actuator model controlled by the AHU supply air temperature controller, and the flow rate through the coil is computed by this network model and returned to the AHU coil thermal model for AHU temperature loop simulation.
other

The model calculates the flow resistance of the coils according to positions of associated valves. The pressure flow and energy characteristics of the constant speed secondary chilled-water pumps are simulated by giving the pump curve and power curve. The flow resistances of the main supply and return pipes and pipes between the zones are considered to be constant. Knowing the number of pumps in operation, the position of AHU moderating valves and the total chilledwater flow rate from chillers, the network model determines the network balance and computes the water flow

Dynamic

simulation

of a building central chilling system

tCcding

Water

Inlet

Cooling Water Outlet

1 1

are the inlet air and water temperatures, C, is the overall thermal capacity of the coil, R, and R2 are the overall heat transfer resistances at the air and water sides. The air and water temperatures at the outlet (t a,ex, tw,ex ) can therefore be computed by the heat balances of both sides:

Fig. 6. Schematic of heat exchanger dynamic model

ta,ex = ta,m _ SHR(ta.in-tc) &Ca

t w,ex
= t w,tn

(19)

tc -

L,m

rates through individual AHUs, the pressures at various points in the network, the flow through individual pumps, the pump power consumption and the water flow rate through the bypass. Primary seawater heat exchanger The dynamic performance of the heat exchangers is represented by one classical steady-state heat transfer model and one simple dynamic model. The heat transfer model computes the number of transfer units (NTU) and heat transfer efficiency (E) using equation (15) for a counterflow heat exchanger and then computes the steady-state seawater and condenser water outlet temperatures. Since the condenser water flow rate is constant in the system, the heat transfer coefficient varies according to the change of seawater flow rate. ,,,(,-2) e=l-e1-+ IndX Two thermal storages are assumed at the seawater outlet and cooling water outlet, respectively, to simulate the dynamic response of a heat exchanger (Fig. 6). Two differential equations represent the dynamic performances as shown by equations (16) and (17):

&Cw

where C, and C, are the capacity flow rates of air and water, SHR is the sensible heat ratio. SHR takes the same value calculated in the same inlet condition in the steadystate case using the bypass factor method. The heat transfer calculation applies the classical number of transfer unit (NTU) and heat transfer effectiveness methods. The classical method to calculate the effect of the fin on the air side on thermal resistance is applied. Two different methods are used to calculate the heat convection coefficient on the air side in dry and wet regions, respectively. In the dry regime, the overall heat transfer resistance (R) is computed as follows:

NT&$=

A Cmi(R,+ Rm+ R,)

Ill,

(21)

(15)
R=

, >

(22)

Tam - Tw,,, 1 Gli Q

w.cd ___

d K,,,e,
dz

Mw,cd(Tw.cd,exP.W

G,cd,exh

(17)

outlet temperatures of seawater and condenser cooling, TX,,,, and are the water outlet temperatures of seawater and T&,cd,ex cooling water after introducing dynamic effects of an exchanger, C,,, and Cw,cd are capacities of thermal storages in the seawater and condenser cooling water sides, respectively. Cooling coil The model is developed on the basis of the mathematical model proposed by Lebrun et al. in Annex 17 [19]. The model includes a steady-state approach and a dynamic approach. A first-order differential equation is used to represent the dynamics of a coil with lumped thermal mass. The dynamic equation on the basis of energy balance ensures that the energy is conserved.

are the steady-state where Sea,,, and Tw,cd,ex

where A is the total heat transfer surface area, R,, R, and R, are the heat transfer resistances of air side convection, coil metal and water side convection, N,,, is the number of the row. In a wet regime, a fictitious air flow is assumed, which has a specific heat equal to the average saturation specific heat (c,, specific heat of saturation moisture air at the average temperature of air inlet wet bulb temperature and water inlet temperature). The air capacity flow rate and air convection coefficient of the fictitious air flow (C,, h,,,) are as follows:

G = m,c,,
ha,wt = ha:, where cP, is the specific heat of moisture air. Then, the overall heat transfer resistance (R) is computed using the same approach: NTU,=C= AU
mm f

A Cm f (R,,, + Rm+ R,)

(26)

or = f

(27)

(18)
where t, is the mean temperature of the coil, fa,inand t,,,,

(28)
(29)

Shengwei

Wang
output computation, control signal output, and waiting time for the next sampling cycle. The PID control function used in DDC loops uses the ISA algorithm, which has the discrete form as shown by equation (34). The handling of the proportional term, integral term and derivative term follows the algorithm implemented in commercial building energy management and control systems. Uk = Kpek+

where subscript wt represents wet regime, f represents fictitious air flow. Having the overall heat transfer resistance, the overall heat transfer resistances in the water side (R,) and air side (R,) are calculated as below. When the coil is in a wet regime, the corresponding values in the wet regime are used.
R = R

I,_, + e(e_,

KAT

+e,)

+KpTdw, I
(34)

&f&P
R,+ R,+R,

R =
2

fLtRnl2
R,+ R,fR;

where (32)
Zk_,+5$(ei_,+ek)=Zk,
I

Actuator, sensor and pipe The actuator model is used to represent the characteristics of actuators [6]. The actuator is assumed to accelerate very quickly and then turn at constant speed. A minimum change (e.g. the sensitivity of the actuator defined as a parameter of the model) in a demanded position is required to restart the actuator. The model includes the hysteresis in the linkage between actuators and valves or dampers. If a valve stem is driven by a rotary actuator, the speed of the valve stem varies with the position of the crank. The actuator model also counts the number of start/ stop or reversals of an actuator and the value of travelled distance of the valve (counted as one unit when a valve moves from its minimum position to its maximum position). These figures provide indications of wear potential of an actuator/valve and hence the cost of failure or maintenance. The dynamics of temperature and pressure sensors are simulated by using the time constant method. One firstorder differential equation (equation (33)) represents the dynamic characteristics of a sensor: dy -=~ dt y-y T, (33)

with ek, ek_, the error signals at current sampling and previous sampling, K, the proportional gain, T, the integral time, Td the derivative time, AT the sampling interval, Zk and Ik_, the integral time at the current and previous time steps, U, the output of PID.

MONITORING OF CHILLING PERFORMANCE

SYSTEM

where y is the true value of the measured variable, 2: is the measured value of the variable, and T, is the time constant. The thermal dynamic characteristics of water pipes are simulated by using the pipe model TYPE3 1 using variable size segments of fluid available in the TRNSYS library. Entering fluid shifts the position of existing segments. The mass of the new segment is equal to the flow rate of the simulation time step. The new segments temperature is that of the incoming fluid. The outlet of this pipe is a collection of the elements pushed out by the inlet flow. Direct digital control Unlike so-called perfect models of a controller normally used to test the energy performance of control strategies, the realistic models developed represent the following functions of the BEMS: Direct Digital Control (DDC) functions, discrete time operation of digital controllers and supervisory control strategies. The time scheduling of a sampling cycle is considered to be in four steps: process variable sampling, control

The existing Building Management System is used to monitor the central chilling system. It is important to calibrate the sensors and transducers of every BMS monitoring point before performance monitoring. Over 2C temperature error in some temperature sensors was observed when calibrating the sensors in this particular building. The performance data monitored by BMS were corrected according to the correlation between the BMS readings and true values obtained by calibration. Evaporator inlet water temperature, evaporator outlet water temperature, condenser inlet water temperature, condenser outlet water temperature, power consumption (kW) of the chillers are recorded every 2 min by the BMS during the chiller monitoring stage. The chilled-water flow rate to the building, supply and return temperatures of chilled water, and bypass water flow rate are also recorded by BMS during monitoring. The water flow rates through evaporators and condensers were measured once and are assumed to be constant. The power consumption of one selected chiller was monitored under various condenser inlet water temperatures (around 27, 30, 31, 32C), chilled-water discharge temperatures (around 7, 8, 9C) and various cooling loads. The performance of other chillers was monitored in certain conditions to check the difference. The performance data in full load were obtained when the chilled-water outlet temperature was obviously higher than its set-point. The performance data in full load and partial load were used to identify the parameters of the chiller using the preprocessor of the chiller model. Figure 7a and b show a comparison between predicted and measured power consumptions and chiller capacities in full load. The difference between the predicted and measured power consumptions is less than 10%. The difference between the predicted and measured capacities in full load is less than 5%. These differences can be explained

Dynamic

simulation

of a building central chilling system

200
Measulod

400
power consumption

600
(kw)

800

(b)
+ 5%. . , 00 / I , /
,

0 Y0

_I o,*

, ,/

/ . , ,, *

9 I

9 09, ,

, -5%

,I

Measwcd

3000 full load capacity

3500 (kw)

Fig. 7. Comparison

between consumption

predicted and and capacity.

measured

power

by the accuracy of the chiller model and the measurement. The heat transfer coefficient of the seawater heat exchanger under various seawater flow rates (Fig. 8a) was obtained by measuring the inlet/outlet temperatures in both sides and the seawater flow rate. The seawater pressure loss across the heat exchangers under various flow rates was measured (Fig. 8b). The pressure loss of the water across the seawater supply and return pipelines under various flow rates was measured (Fig. 8~). The pressure head and power consumption of one selected seawater pump under different flow rates and different input frequencies (35, 40, 45, 49Hz) were monitored. The other pumps were tested in certain selected conditions. No significant differences between different pumps were observed. The secondary chilled-water distribution pump performances and pressure losses on the AHLJ including valves were also tested in a few conditions. The power consumptions of the constant speed condenser and evaporator water circulation pumps were measured manually and are assumed to be constant. SYSTEM SIMULATION AND EVALUATION

A Transient System is used as the platform

Simulation Program (TRNSYS) for the dynamic simulation of a

HVAC system. Figure 9 shows the schematics of the interconnection of the component models (information diagram) in simulation. Five AHUs are simulated, each of which represents the dynamic load of AHUs in one of the five zones of the entire building. Only one zone is occupied 24 h a day. The AHU outlet temperature is controlled by a digital PID controller at a given set-point which is different from season to season. A two-way moderating valve is used to control the water flow rate through a coil, which is driven by an actuator according to the control signal given from the temperature controller. The flow rate through each simulated AHU coil is multiplied by a factor to represent the total load of all AHUs in the zone. To avoid the difficulty in convergence when solving the pressure flow balances of a chilled-water network and seawater network, the pressure flow balances of the chilled-water network and seawater network are simulated by separate models. The water flow rates through cooling coils and heat exchangers as the outputs of the water network models are given to the performance models as inputs. The temperature controller will adjust the valve positions and then the flow rate to control the temperatures at the set-points. The air flow rate and AHU inlet temperatures for different seasons are given by data files during simulation as a test condition, which has slightly different patterns for the AHUs in different zones to ensure that control actions of AHUs in different zones are not simultaneous. The overall cooling load of the AHU in the entire building in four different seasons used in the simulation exercises is shown in Fig. 10, which is selected by referring to the actual building load in different seasons from field monitoring. Tests were performed to check the accuracy of the system performance obtained by simulation. Tests show that the performance data given by simulation generally match the measured values in the same condition. The comparison of some important variables of a 1 day test are shown in Fig. 11. The simulation was conducted by giving the monitored data of numbers of chillers and pumps in operation, chilled-water supply and return temperatures, seawater pump frequency and seawater temperature as test conditions. Figure lla shows the total consumption of chillers. One chiller was used overnight and two chillers were used during the day. The difference between the simulated and measured chiller total power consumption is within 50 kW and not more than 10% in the test, with exception for a very short period. Figure 11 b shows the comparison between simulated and measured condenser inlet and outlet temperatures in the same test case. The simulated condenser inlet temperature was slightly higher than the measured values (about 0.4 to 0.8C) at most times in the day. The simulated condenser outlet temperature was about 0.5 to 1.2C higher than the measured values at most times of the day. Figure 1 lc and d show the comparison between the simulated and measured seawater flow rates and pump consumptions. The flow rate was monitored by the BMS and the pump power consumption was recorded manually. The difference between the simulated and measured seawater flow rates was not more than 4% in the test. The

10

Shengwei

Wang

80

100

120

140

160

180

200

220

Water Flow
300

Rate (L/s)

250

200

6 150 D e SL 100

50

0 100 125 150 175 Water Flow Rate (L/s) 180 160 140 120 3 2 5 s L 100 80 60 40 20 0 200 225 250

L
300 350 400 450 500 550 Water Flow Rate (L/s) Fig. 8. Performance of heat exchanger and seawater pipes,

: TXX: component Taoset. T: temperature M: flow rate W: power eonsumtion Num.p: number of pump in operation SUBSCRIPT a: air c: chiller i: in to unit set: set point
Valve Valve Valve +

lgpe number rnodrl P: pressure

LOAD,Ma, T~I of each ZONE

Hyd: hydraulic

P: pump w: water 0: out from unit ch: chiiled water

Two Position I/ II 11 Two Position Position

I: / ~Two

Valve Position

w Two

Chiller Power Consumption

WC

a
P SENSER Num.p Fig. 9. Schematic of TRNSYS simulation DECK.

z 2

Tm, UA, dUA/dP, Wp, dWp/dP

12

Shengwei

Wang

8000 E = d 4 4000

6000

1 -

Winter Middle-l Middle-II

~ !

3:oo

6:00

9:oo

12:oo

1500

18:OO

21:oo

0:oo

Time (hour) Fig. 10. Total cooling load of AHUs in the building.

difference between the measured and simulated seawater pump power consumption was less than 3% in the test.

CENTRAL

CHILLING SYSTEM CONTROL STRATEGIES

Sequence and local loop controls Three local loop control strategies and one supervisory control strategy are used to control the chilling system. Five PID controllers are used to control the AHU supply air temperatures. The chiller sequence controller controls the number of chillers in operation according to a few criteria, increasing the number of chillers and reducing the number of chillers. If a negative bypass Aows over a certain value for significant time, one more chiller will be switched on. If the measured total chilling load of the building is, for significant time, less than the total cooling capacity of the running chillers when one of the currently running chillers is switched off, one chiller will be switched off. The number of seawater pumps in operation depends on the number of chillers in use, i.e. one pump is used if one or two chillers are in operation, two pumps are used if three or four chillers are used. The sequence controller of the secondary pumps determines the number of pumps in use according to the measured pressure head at the secondary pump supply. Switch set-points for increasing the number of pumps (from one to two and from two to three) and reducing the number of pumps (from three to two and from two to one) are set as the parameters of the controller. One PID pressure controller is used to maintain the seawater pressure head at the inlet of the heat exchangers at its set-point, which is reset by the seawater pressure supervisory control strategy. The pressure controller senses the pressure at the inlet of heat exchangers and adjusts the frequency output of the frequency inverters using the PID algorithm.

Optimal pressure set-point reset On-line optimal reset of the seawater pressure set-point is achieved by a derivative control strategy and an adaptive control strategy (Fig. 12). The derivative control strategy resets the pressure set-point by adjusting the pressure set-point according to the estimated derivative of the total power with respect to seawater pressure. The adaptive strategy is used to identify the system parameters essential for the derivative control strategy. The derivative control adapts the knowledge of operators in supervising and optimising a controlled variable. The controller estimates the derivative of total power W,,, with respect to the seawater pressure (P) across the heat exchanger. If the value of the derivative is positive, the pressure should be reduced. In contrast, the pressure should be increased if the derivative has a negative value. The pressure will approach the optimal point when the derivative approaches zero at a certain building load or cooling load. In the actual operation of a building chilling system, the adjustment will be a regular task for a controller to maintain a zero derivative value, since the load is changing. The BMS estimates the derivative of the pump power on the basis of equation (35) using the system information available from conventional BMS monitoring systems. To properly adjust the pressure, the controller should identify the derivatives of pump power and the heat transfer coefficient with respect to the pressure, and should obtain the chiller energy power, mean temperature difference and heat transfer coefficient. (35) where C, is a denser is the cooling W,, is seawater pump power, WCis chiller power, constant coefficient representing the effect of concooling water temperature on chiller power, T, mean temperature difference between condenser water and seawater in an exchanger.

Dynamic

simulation

of a building central chilling system

13

1400

1200

1000

0 g 800

E z s

600

v B g a

a
-+-Power -Power (Simulated) (Measured)

400

200

Time (hour)

40

@I
35

30 5 2 25 e E g 20

E F

1.5
Temp. at Condensor Outlet (Simulated) 10 -+ Temp. at Condensor Inlet (Measured) at Condensor
Outlet (Measured)

-+Temp. 5

Time (hour)

14

Shengwei

Wang

500 450 400 350 z 2 300 Q) 3 250 ti 4 200 IA 150 100 50 0

( cc>

-Total

Sea Water Flow Rate (Simulated) 1

Time (hour)

100

Seawater Pump Consumption (Simulated) (Measured)

-t Seawater Pump Consumption L__

Dynamic
Total Pump Power

simulation

of a building central chilling system

P. Tseal. (Hz) ~... 4

15

Power of Chillers

TW.Cdl TW.CdZ Tsea2 (on/off)

TW,CXII Tw.cd2 Tseaz

(on/off)

Tw,cdl 1 Tw,cd2 - Tsea2 - (on/off)

Fig.

12.

Schematic of pressure set-point reset strategies

Having the derivative of the total power with respect to the pressure, the controller adjusts the pressure setpoint with a change rate determined by the derivative function shown in equation (36): (36) where dP,,,/dt is the change rate for pressure set-point adjustment and K is the proportional gain (a positive constant). Besides the chiller power, heat transfer coefficient and mean temperature difference, two derivatives (derivatives of heat transfer coefficient and pump power with respect to pressure) are essential in order to identify the derivative of total power. Recursive Least-Squares (RLS) estimation with exponential forgetting is implemented to identify the derivatives of the heat transfer coefficient and pump power with respect to the pressure. It is also used as a filter in identifying the heat transfer coefficient. Secondorder models for the pump power and heat transfer coefficient are used, which are shown by equations (37) and (38): UA(t) = Wp(O =

power, and the heat transfer coefficient, using the coefficients identified.

are estimated

PERFORMANCE TEST AND EVALUATION OF CONTROL STRATEGIES Two simulation exercises were conducted to test the dynamic response and energy performance of the system and control strategies under various loads. In the first exercise, three 1 day tests were performed in each of the four cases. In the first test, the pressure is reset by the optimal reset strategy. In the other two tests, the pump was set to be minimum or maximum speed. In the second exercise, the control strategies were tested over 4days with the same building cooling load of the summer case for all four days. The heat transfer coefficient of the heat exchangers reduced gradually for 30% over 4 days. Figure 13 presents examples of pressure control loop stability during tuning the parameters of the controllers. Figure 13a shows the performance of seawater pressure control loop when an unsuitable (too high) proportional gain of the PID pressure controller was used. The frequency of the pump was not controlled stably and the pressure across the heat exchangers and pumps was oscillating. The pressure set-point oscillated slightly. Figure 13b shows the performance of the same control loop when an unsuitable proportional gain of pressure setpoint reset controller was used. The pressure set-point oscillated obviously and caused the oscillation of the pump frequency and pressures across the heat exchangers and pumps. The pressure was significantly higher than its set-point during the night since the pump frequency was at a minimum and cannot be reduced further.

MfMl(or~2(t)l *Lp(o>m21~ Mf),c,(o,G(~)l* LpmYo21T,

(37) (38)

where b,(t), b,(t), b,(t), co(t), c,(t), c2(t) are time-dependent parameters to be identified. Four estimators for heat transfer coefficient are used, one for each chiller exchanger. Each estimator requests the water temperatures at the inlet and outlet of the relevant exchanger, and the pressure. One estimator is used for pump power derivative. At each sampling step, the derivatives of heat transfer coefficient and pump

16
280

Shengwei

Wang

Pressure Head of Pump

240

n ,d

200

80

40

Pressure Set-Point

0 0 2 4 6 8 10 12 14 16 18 20 22 24

20

Time (hour)

280

60 55 50 45 Q z z? 40 9 % 35 % l Pressure across Exchanger Pressure Set-Point 30 25 20

240

0 _
0

-7

10

12

14

16

18

20

22

24

Time (hour)
Fig. 13. Examples of seawater pressure control loop stability during tuning.

Figure 14 shows the seawater loop and chiller operation during tests on optimal reset strategies in different cases. Figure 14a shows the pressures and seawater pump frequency of optimal reset strategy in summer case. Figure 14c shows the number of chillers in operation in the test of summer and middle-II cases. In night operation mode, the pump was set to a minimum speed, whilst only one chiller was in operation. The pressure

frequency had a peak at the beginning of the day operation mode, which was the result of high start chiller load and the response of the strategy to the sudden load change. It disappeared quickly, which shows that the stability of the strategy is acceptable. During the main part of the day, the reset controller regulated the pressure according to the change of the building cooling load, while four chillers were in operation. It can be observed
and

Dynamic simulation of a building central chilling system

300 60

17

250

200

!,,, & I k 100

uh 40 g s 35%

30 50 25

0 0 2 4 6 8 10 12 14 16 18 20 22

-t
24

20

Time (hour)
280

(b)
Pressure Head of pump

60

240

55

50

200

3 2 t E I k 80 120 160

Pressure

across Exchanger

40

25

0 0 2 4 6 8 10 12 14 16 18 20 22 24

20

Time (hour)

4-

3-; & -z 2 2i

I I / -_l- I

-1~

~__

_-_t+.__

-7~-.....-~-

,___~---i--___~__.

~~_

10

12 Time (hour)

14

16

18

20

22

24

Fig. 14. Seawater

pressures,

pump frequency

and chiller operation

(exercise

1)

18

Shengwei
Table 1, Electricity

Wang
at various loads Chiller + seawater pump ( 10h kJ) 48.21 55.11 48.21 87.0 1 87.79 82.75 118.74 123.78 116.75 155.29 155.06 149.70 Comparison with adaptive reset (%) 0.00 + 14.31

consumption

Test case Winter (1 day)

Control

strategy

Seawater pump (lo6 kJ) 6.31 16.39 6.44 6.30 16.20 8.60 9.00 23.42 11.16 9.29 24.19 13.34

Chiller ( 10h kJ) 41.90 38.72 41.77 80.71 71.59 74.14 109.74 100.36 105.59 146.00 130.88 136.36

Minimum Maximum Adaptive reset Minimum Maximum Adaptive reset Minimum Maximum Adaptive reset Minimum Maximum Adaptive reset

Middle-l

(1 day)

f5.15 +6.10 _ fl.70 + 6.02

Middle-II

(1 day)

Summer

(1 day)

f3.83 +3.68

that the pressure changed following the change of chiller load. A peak appeared around 16:OOpm with the chiller load. Near the end of the day, the controller set the pressure at a low level, since two chillers worked at partial load. Figure 14b shows the pressures and seawater pump frequency of optimal reset strategy for middle-II case. In night operation mode, the pump was at minimum speed. During the main part of the day, the reset controller regulated the pressure according to the change of the building cooling load, while three chillers were in operation. Around 16:00pm, there was an obvious increase of pump frequency due to the peak of the chilling load. The two shocks occurring were the response of the control to the changes of chiller number between three and four. Before the end of the day, two chillers worked at high load, the controller set a high frequency for the single pump in operation. Energy statistics are shown in Table 1. In the summer case, the optimal pressure set-point reset strategy saved 3.83% of the total electricity in comparison with the electricity consumption when setting the pump at minimum speed, and saved 3.68% of the total electricity in comparison with the electricity consumption when setting the pump at maximum speed. In the middle-II case, it saved 1.7% of the total electricity in comparison with the electricity consumption when setting the pump at minimum speed, and saved 6.02% of the total electricity in comparison with the electricity consumption when setting the pump at maximum speed. In the middle-I case, it saved 5.15% of the total electricity in comparison with the electricity consumption when setting the pump at minimum speed, and saved 6.1% of the total electricity in comparison with the electricity consumption when setting the pump at maximum speed. In the winter case, it had no saving
Table 2. Electricity Control strategy (4days with variable (/A) Minimum Maximum Adaptive reset consumption

on the total electricity in comparison with the electricity consumption when setting the pump at minimum speed since the optimal control was very close to the minimum setting. But it saved 14.31% of the total electricity in comparison with the electricity consumption when setting the pump at maximum speed. Figure 15 shows the pressure across the heat exchangers and pump, and the frequency for the seawater pumps in the second exercise. Similar patterns of the controlled pressure and frequency curves during 4days show that the control strategies have consistent performance. The controlled pressure and frequency increased gradually over 4 days. It shows that the optimal control set-point increased as the heat transfer coefficient reduced. The energy performance of the control strategies in the exercise is shown in Table 2. The adaptive pressure setpoint control strategies saved 5.31% of the total electricity consumption in comparison with the electricity consumption when setting the pump at minimum speed, and saved 3.11% of the total electricity in comparison with the electricity consumption when setting the pump at maximum speed. When comparing the data with that in the summer case test in the first exercise, the optimal point of the pressure setting was closer to the maximum pump speed and the daily total electricity consumption was 2.4% higher, due to the reduction of the heat transfer coefficient of the heat exchangers.

SUMMARY

AND CONCLUSIONS

Simulation exercises show that the dynamic system simulation is a convenient and suitable tool in testing and evaluating the control performance of large central chilling systems and their on-line control strategies. However, simulation tests show that there are difficulties in
with variable system characteristics Chiller (lOh kJ) 608.57 535.44 555.47 Chiller + seawater pump (IO6 kJ) 645.72 632.23 613.15 Comparison with adaptive reset (%) +5.31 f3.11

Test case Summer

Seawater pump ( lo6 kJ) 37.15 96.79 57.68

Dynamic simulation of a building central chilling system

19

Pressure Head of Pump

Pressure cross Exchanger

I 20 0 6 12 18 24 30 36 42 48 54 60 66 72 78 84 90 96

Time (hour) Fig. 15. Seawater pressures, pump frequency and chiller operation (exercise 2)

convergence when component models representing both the thermodynamic and hydraulic performances of individual components are integrated to simulate the realistic (i.e. thermodynamic, hydraulic, etc.) performances using component-based simulation programs such as TRNSYS. The use of separate models to handle the pressure flow balance of entire water networks is one solution for such a problem. The test results also show that significant energy saving in seawater-cooled central chilling systems can be

achieved by properly resetting the pressure set-point and the seawater pump speed using on-line optimal control strategies. These on-line control strategies can be tested, commissioned and pretuned by using them to control the living chilling systems in simulation. Acknowledgement-This paper presents the contribution of the author to the IEA (BCS) joint research project Annex 30 (Bringing Simulation into Application). The research work in The Hong Kong Polytechnic University is financially supported by a University Research Grant.

REFERENCES 1. Kelley, G. E., Control system simulation in North America. Energy and Buildings, 1988, 10, 1933202. 2. Dexter, A. L., Control system simulation - computer control. Energy and Buildings, 1988, 10, 203211. 3. 4.

5.
6.

M. and Goh, P. A., Transient response of a closed-loop VAV system. ASHRAE 1991, 93(2), 78. Maxwell, G. M., Shapio, H. N. and Westra, D. G., Dynamics and control of a chilled water coil. ASHRAE Transactions, 1989, 95(l), 1243. Dexter, A. L. and Haves, P., A robust self-tuning controller for HVAC applications. ASHRAE
Transacrions, Transactions, 1989, 95(2), 431438.

Zaheer-Uddin,

7.
8. 9.

10. 11. 12. 13. 14.

Haves, P. and Dexter, A. L., Simulation of local loop controls. In Proceedings ofBuilding 89. IBPSA, Vancouver, 1989. Zaheer-Uddin, M., Optimal, sub-optimal and adaptive control methods for the design of temperature controllers for intelligent buildings. Building and Environment, 1993, 28(3), 311-322. Zaheer-Uddin, M., Temperature control of multizone indoor spaces based on forecast and actual loads. Building and Environment, 1994,29(4), 485-483. Wang, S. W. and Burnett, J., BEMS control strategies: Evaluation of realistic performance by computer simulation. Building Services Engineering Research and Technology, 1996, 17(l), 15-19. Lebrun, J. and Wang, S. W., Evaluation and emulation of building energy management systems synthesis report. IEA (BCS) Annex 17 Final Report, University of Liege, Belgium, 1993. Madjidi, M., Simulation exercises A.2 - residential heating system. IEA (BCS) Annex 10 Final Report, University of Stuttgart, Germany, 1991. Building simulation exercisessair-conditioning system. IEA (BCS) Annex 10 Final Report, University of Liege, Belgium, 1991. Carey, C. W., Mitchell, J. W. and Beckman, W. A., The control of ice storage systems. ASHRAE Journal, May 1995, 37(5), 32-39. Karki, S., Wang, S. W., Peitsman, H. C., Kelley, G. E., Haves, P. et al., Development of Emulation Methods. IEA (BCS) Annex 17 Publication, Finland, 1993.

20
15. 16.

Shengwei Wang
Wang, S. W., Haves, P. and Nusgens, P., Design, construction and commissioning of building emulators for EMCS applications. ASHRAE Transacrions, 1994, 100(l), 146551473. Kelly, G., Park, K. and Barnett, J. P., Using emulator/testers for commissioning EMCS software, operator training, algorithm development, and tuning local loops. ASHRAE Transactions, 1991, 97(l), 6699678. Haves, P. and Dexter, A. L., Use of a building emulator to emulate control strategies implemented. In Commercial BEMS, Proceedings of Building Environmental Performance 91, Canterbury, 1991. Peitsman, H., Wang, S. W., Haves, P., Karki, S. and Park, C., The reproductivity of tests on energy management and control systems using building emulators. ASHRAE Transactions, 1994, 100(l), 1455-1464. Lebrun, J., Ding, X., Eppe, J. P. and Wasacz, T., Cooling coil models to be used in transient and/or wet regimes - theoretical analysis and experimental validation. In Proceedings of System Simulation in Buildings 90, Liege, Belgium, 1990.

17. 18.

19.