Sunteți pe pagina 1din 12

The 7th Greater Pearl River Delta Conference on Building Operation and Maintenance

A hybrid control for chiller sequencing control with


enhanced robustness under uncertainties
Y. D. Liao (1), G.S. Huang (2)
1. ydliao2-c@cityu.edu.hk
2. gongsheng.huang@cityu.edu.hk
(1), (2) Department of Architecture and Civil Engineering, City University of Hong Kong,
Hong Kong.

Abstract

Chiller sequencing control is significant for the operational feasibility and


energy-efficiency of multiple-chiller plant. Various control strategies have been
developed and applied to achieve overall energy efficiency of the chiller plant while
not scarify the cooling sufficiency for indoor thermal comfort. However, these control
strategies suffer from various uncertainties and usually fail to achieve the expected
control and/or energy performance. Therefore, this study proposes a hybrid control to
enhance the robustness and reliability of chiller sequencing control under uncertainties.
The proposed control hybridizes two conventional control strategies, viz. the chilled
water return temperature-based control and the direct power-based control, by making
best use of their complementary advantages under uncertainties. Using historical
operation data, the robustness and performance of the proposed hybrid control is
analyzed and compared with conventional controls under multi-levels of uncertainties.
Robustness analysis and comparison studies show that the proposed hybrid control is
able to achieve better robustness when various uncertainties exist.

Keywords

Chiller sequencing control; uncertainty; hybrid control; robustness enhancement.

1 Introduction

A centralized chiller plant is the core equipment in a central air-conditioning system,


providing cooling for indoor space so as to achieve indoor thermal comfort during hot

93
The 7th Greater Pearl River Delta Conference on Building Operation and Maintenance

seasons. In order to improve its energy efficient operation at various load conditions,
the chiller plant is always configured with multiple chillers connected in parallel. In
this configuration, chiller sequencing control is an essential function, which is used to
switch chillers on or off according to building instantaneous cooling load, aiming at
achieving an overall coefficient of performance (COP) of the online chillers as high as
possible while fulfilling the demanded cooling load [1,2].

Various chiller sequencing control strategies have been developed and implemented in
practice. Basically, there are four typical control strategies, including total cooling load
based (Q-based) sequencing control, chilled water return temperature based (T-based)
sequencing control, the direct power based (P-based) sequencing control and bypass
flow rate based (F-based) sequencing control [3]. Different control strategies use
different load indicators to represent the building cooling load and they are expected to
switch chillers on/off economically.

However, those load indicators may not accurately reflect the real load condition for
the reason that potential problems may exist in measurement tools or local loop
controls [4, 5]. Hence, these control strategies suffer from uncertainties and the control
performances may deviate from the expectation. The study of Liao et al. [6,7] shows
that these uncertainties may cause not only delayed on/off switches but also
unnecessary frequent on/off switches. The delayed switches will lead to energy waste
or insufficient cooling supply; while the frequent switches may cause system instable
and thus shorten the service life of equipment. Uncertainty is one of the main reasons
why the automatic chiller sequencing control in real applications was often switched to
manual control [8].

In order to promote the application of automatic chiller sequencing control, it is


important to enhance the robustness of the chiller sequencing control under
uncertainties. To this end, several methods have been proposed. For example, Huang et
al. [9] proposed a data fusion technique to fuse the redundant measurements so as to
reduce the measurement uncertainty in the estimation of building cooling load. Li et al.
[10] use stochastic probability theory to determine the switch actions in a statistical
way, so that the impact of uncertainty on individual measurement is attenuated.

The methods reviewed above focused on the Q-based control, which uses the measured
cooling load as the load indicator. However, in current practice the other three are also
commonly used due to their simplicity (only one measurement is used for load
indicator) and their own advantages. Therefore, in this study, a hybrid control based on
two typical control strategies is developed to improve the control performances of
chiller sequencing control under uncertainties. In the proposed control, the T-based
control and the P-based control are hybridized to determine the sequencing operation,
i.e. using the T-based control to determine switch-on action and using the P-based
control to determine the switch-off action; while simultaneously the power of online
chiller is used to correct the switch-on actions to further enhance the reliability of the
chillers’ switch-on control.

The rest of this paper is organized as follows. Section 2 introduces the characteristics

94
The 7th Greater Pearl River Delta Conference on Building Operation and Maintenance

of the two control strategies under uncertainties. Section 3 describes the methodology
of the hybrid control and the associated uncertainties. Section 4 presents robustness
analysis and performance comparisons. Finally, conclusions are given in Section 5.

2 Chiller sequencing control strategies and associated uncertainties

2.1 Characteristic of the T-based sequencing control

The T-based control uses the chilled water return (CHWR) temperature at the header
pipe as an indicator of the cooling load and switches chillers on/off by comparing the
CHWR temperature with pre-defined thresholds. One advantage of the T-based control
is that there is only one variable that needs to measure, which makes it easier to be
implemented and suffer less from measurement uncertainty. The other advantage is that
the T-based control has better reliability in cooling supply under uncertainties [6]. This
is because the CHWR temperature is sensitive to the load variation. When the cooling
supply is not (even slightly not) sufficient, the chilled water return temperature will
increase quickly, so that the T-based control is less risk to have severe cooling
insufficiency under uncertainties.

However, uncertainty analysis shows that the T-based control suffers from three types
of uncertainties simultaneously: viz. measurement uncertainty from the sensor
inaccuracy; control uncertainty in the chilled water supply temperature control and
operational uncertainty from the constant pumps in the primary side [6]. These
uncertainties may lead to periodical fluctuations in CHWR temperature and result in
unnecessary switching actions, especially when approaching to switch-off condition as
under this situation the CHWR temperature has a low value. Unnecessary switching
will jeopardize system stability and may shorten the service life of chiller plant.

Based on the discussion above, it is found that the T-based control is more suitable for
switching chillers on rather than switching chiller off.

2.2 Characteristic of the P-based sequencing control

The P-based control is to switch chillers on/off according to the loading condition of
the online chillers. The cooling load is indicated by the percentage of full-load
amperage (PFLA) of the online chillers’ compressor power. When the PFLA of online
chillers is larger than predefined switch-on thresholds for a certain period, one chiller
will be switched on; while when the PFLA of online chillers is smaller than predefined
switch-off threshold for a certain period, one chiller will be switched off.

The main advantage of the P-based control is that the measurement of PFLA is a very
reliable variable since the amperage measurement has higher accuracy and it is
commonly available on chiller control panel [11]. However, the PFLA is only a
secondary data source of cooling load and it should be correlated to cooling load for
setting up switch-on/off thresholds. The thresholds are normally setup according to the
rated COP curve of the chillers while the actual COP curve always deviates from the

95
The 7th Greater Pearl River Delta Conference on Building Operation and Maintenance

rated one. The threshold uncertainty exists in the P-based control. Uncertainty analysis
shows that the COP error can result in a deviation in the calculated PFLA. The
switch-on thresholds may not be reached due to this deviation and consequently, no
extra chiller is called on when an extra chiller is really needed [6,7]. Uncertainty
analysis also shows that the threshold uncertainty has less impact on the switch-off
control than on the switch-on control. This is because the switch-off thresholds have
relatively low values and thus the absolute deviation between the pre-defined
switch-off thresholds (set according to the rated COP curve) and the ideal switch-off
thresholds (should be set according to the actual COP curve) is smaller.

Based on the discussion above, it is found that the P-based control is more suitable for
switching chillers off rather than switching them on under uncertainties.

3 Hybrid control
3.1 Overview of the hybrid control

As discussed in Section 2, the T-based control is more suitable for switch-on control
rather than switch-off control; while the P-based control is more reliable for switch-off
control rather than switch-on control. Considering the complementary characteristics
of them, a hybrid control is developed to enhance the robustness of chiller sequencing
control, viz. to use the T-based control to determine the chillers’ switch-on actions and
to use the P-based control to control the chillers’ switch-off actions; while the power
condition of online chillers is simultaneously used to correct the switch-on actions so
to further reduce the unnecessary switch-on actions. The switch-on/off criteria are
given below:

 Switch-on criterion: IF the chilled water return temperature is larger than a


pre-defined switch-on threshold and this state lasts for a period longer than the time
limit (ton,l), and at the same time the power of online chillers is larger than a
pre-defined threshold, THEN a chiller and its interlocked pump will be switched
on.
 Switch-off criterion: IF the PFLA of online chillers is smaller than the pre-defined
switch-off threshold and this state lasts for a period larger than the time limit (toff,l),
THEN a chiller and its interlocked pump will be switched off.
𝑜𝑛
The flowchart of the hybrid control is illustrated in Figure 1, where 𝑃𝑧𝑡ℎ,𝑟 is the
subsidiary criterion used to correct the switch-on control by identifying the loaded
condition of the online chillers. If the online chillers are approaching the full-load
condition, the switch-on command of the T-based control is deemed to be valid and
one extra chiller will be called on. If the online chillers are not approaching full-load
condition, the switch-on command of the T-based control is judged as invalid and no
chiller will be called on. This subsidiary criterion is set up to correct the unnecessary
actions of the T-based control under the impacts of uncertainties.

96
The 7th Greater Pearl River Delta Conference on Building Operation and Maintenance

Qload(k)

Trtn,meas(k) Pmeas(k)

Trtn (k ) = Trtn ,meas (k )


P(k ) = Pavg (k )
PFLA(k ) = I meas (k ) / I z , full −loaded Load indicator

Trtn (k ) > Tzth


on
?
ton > ton ,l ? PFLA(k ) < PFLAzth off
?
Yes P(k ) > P ? on toff > toff ,l ? Yes
z

No No
Switch on one Keep the Switch off one
extra chiller current state, online chiller
N(k)=N(k-1)+1 N(k)=N(k-1) N(k)=N(k-1)-1

Online chiller
Sequencing
N(k)
control

Figure 1 - Control logic of the hybrid control

The threshold of the subsidiary criterion is set according to the rated switch-on
threshold of the P-based control plus the maximum COP error range (which can be
obtained from site operation data) as follows
on
Pzth (1)
P on
= ,r
z
∆ COP
1+ ( ) max
COPr


𝐶𝑂𝑃 𝑜𝑛
where, (𝐶𝑂𝑃 )𝑚𝑎𝑥 is the maximum range of the COP error at thresholds; 𝑃𝑧𝑡ℎ,𝑟 is the
𝑟
th
z switch-on threshold that set according to the rated COP curve. To reduce the
measurement uncertainty in individual measurement of chiller power, P(k) is the
average value of a few continuous measurements (normally 2 or 3 samples).

3.2 Uncertainties associated with the hybrid control

Since the T-based control and the P-based control are hybridized, the uncertainties
associated with the two control strategies are suffered by the hybrid control. The
uncertainties suffered by the switch-on control include the control uncertainty,
operational uncertainty and measurement uncertainty. The uncertainties associated with
the switch-off control are the threshold uncertainty and measurement uncertainty. Use
the uncertainty shifting method developed in Liao et al. [6] to shift all these
𝑜𝑛
uncertainties to the load indicators 𝑇𝑟𝑡𝑛 and I used in the hybrid control, the total
uncertainties associated with the hybrid control can be described by Eqn. (2) and Eqn.
(3).

97
The 7th Greater Pearl River Delta Conference on Building Operation and Maintenance

∆ m, p ,oper  ∆ 
∆T ,rtn = (Trtn,meas − ∆T ,rtn,meas − Tsup,st ) − ∆T ,sup,cont 1 + m, p ,oper  − ∆T ,rtn ,meas

(2)
m p ,r  m p ,r 

∆ COP
∆ I , z = ∆ I ,meas −
COPr
( I − ∆ I ,meas ) (3)

where, ΔI,meas , ΔT,rtn,meas is the measurement uncertainty associated with the measured
PFLA and the CHWR temperature, separately; Δm,p,oper is the operational uncertainty;
ΔT,sup,cont is the control uncertainty and ΔCOP is the COP error contributing to threshold
uncertainty.

3.3 Uncertainty modeling

The measurement uncertainties have random property and they can be described using
normal distributions [12]. Since the bias in measurements is not considered in this
study, the measurement uncertainties ΔT,rtn,meas and ΔI,meas are modeled by

∆ ~ N 0,σ 2 ( ) (4)

where σ is the standard deviation and can be estimated from site operation data or from
the sensor marked accuracy E (given by manufacture).

The analysis of chiller operation data in Liao et al. [6] also shows that the control
uncertainty and operation uncertainty have the pattern of periodic variations and can be
described by a sine function as follows
nc

∑ Ai sin (ωit )
∆ con ,oper =
i =1
(5)

where nc is the number of components that should be taken into account; ωi , ci are the
frequency and amplitude of ith component, respectively. According to the site operation
data analysis presented in Liao et al. [6], the threshold uncertainty associated the
switch-off control can also be described using a normal distribution. The standard
deviation can be estimated from the site operation data by a statistic method.

4 Case studies
4.1 Simulation platform

In order to evaluate the improved performance of the hybrid control, robustness


analysis was firstly conducted to the hybrid control with considering multiple levels of
uncertainties. Then the performance of the hybrid control was compared with two
typical control strategies (viz. the T-based control and the P-based control) under the
same uncertainty conditions.

To perform the robustness analysis and the performance comparison, a simulation


platform of a typical decoupled multiple-chiller system with structure as shown in
Figure 2 was constructed using TRNSYS 17 [13]. In the primary loop, four identical

98
The 7th Greater Pearl River Delta Conference on Building Operation and Maintenance

600kW centrifugal chillers were parallelly connected and four constant speed pumps
with a rate flow rate 28.65L/s were interlocked with the four chillers. The chilled water
from chillers was mixed at header pipe with a temperature setpoint of 7oC for all
chillers. In the secondary loop, a variable-speed pump with a rated flow rate 114.6L/s
was installed to deliver the chilled water to air-handling units (AHUs). The pump
speed was controlled to maintain the constant differential pressure at the critical branch.
The supply air temperature from AHUs was controlled by a PID controller through
varying the opening of the valves on chilled water pipes and the flow rate of supply air
was varied according to the cooling load variation. One-week cooling load collected
from a high rise building was used in the study [6]. The sequencing control strategy
was coded using MATLAB and communicated with TRNSYS 17 through the exiting
interface, i.e. Type 155.

Figure 2 - The structure of a typical decoupled multiple-chiller system

The thresholds used in the three controls were listed in Table 1. Note that the Q-based
control was considered as the benchmark for performance comparison since it directly
use the measured cooling load as the indicator of chiller sequencing control. The
thresholds of the Q-based control were also considered as the basis for the thresholds
setup of the other control strategies. The switch-on and -off criteria needed to be met
for continuous 10 minutes and then the associated switch on and off actions were taken.
Such 10-minute time counting was mainly used to avoid short-time switching in
practice [14].

Table 1 - Thresholds used in chiller sequencing control strategies


Control strategy Switch-on thresholds Switch-off Thresholds
From 1 to 2: Trtn > 11.94 oC From 2 to 1: PFLA < 45.60%
Hybrid control From 2 to 3: Trtn > 11.85 oC From 3 to 2: PFLA < 62.07%
From 3 to 4: Trtn > 11.81 oC From 4 to 3: PFLA < 70.30%
From 1 to 2: PFLA > 98.13% From 2 to 1: PFLA < 45.60%
P-based control From 2 to 3: PFLA > 95.34% From 3 to 2: PFLA < 62.07%
From 3 to 4: PFLA > 94.45% From 4 to 3: PFLA < 70.30%
From 1 to 2: Trtn > 11.94 oC From 2 to 1: Trtn < 9.28 oC
T-based control From 2 to 3: Trtn > 11.85 oC From 3 to 2: Trtn < 10.10 oC
From 3 to 4: Trtn > 11.81 oC From 4 to 3: Trtn < 10.51 oC
From 1 to 2: Q > 592.8 kW From 2 to 1: Q < 547.2 kW
Q-based control From 2 to 3: Q > 1162.8 kW From 3 to 2: Q < 1117.2 kW
From 3 to 4: Q > 1732.8 kW From 4 to 3: Q < 1687.2 kW

99
The 7th Greater Pearl River Delta Conference on Building Operation and Maintenance

The multiple levels of uncertainty setup for robustness analysis and performance
comparisons are listed in Table 2. Level 1 represents design condition of the chiller
plant in which all the values are recommended by design guides [15]. Level 5 indicates
the poor working conditions with uncertainties extracted from site operation data. The
Level 2 to Level 4 have the uncertainties between Level 1 and Level 5 using linear
interpolation. It should be noted that for measurement uncertainty, the values given in
table represents the accuracy of sensors. For the control uncertainty and operational
uncertainty, two components of the sine function were selected. For the threshold
uncertainty, ∆𝑐𝑜𝑝 /𝐶𝑂𝑃𝑟 is the error range of COP at thresholds.

Table 2 - Uncertainty levels


Measurement uncertainty Control uncertainty Operational uncertainty Threshold
Temperature Flow Current uncertainty
Levels
sensor meter meter
Magnitudes Frequency Magnitudes Frequency ∆ COP / COPr
(oC) (Hz) (%) (Hz)
(oC) (%) (%) (%)
Level 1 ±0.1 ±1.0 ±1.0 0.5/0.5 0.6/0.4 ±1.0
Level 2 ±0.2 ±1.5 ±1.5 1.5/1.5 1.8/1.2 ±2.0
2.34*10-4/ 2.0*10-4/
Level 3 ±0.3 ±2.0 ±2.0 2.5/2.5 3.0/2.0 ±3.0
1.8*10-4 2.6*10-4
Level 4 ±0.4 ±2.5 ±2.5 3.5/3.5 4.2/2.8 ±4.0
Level 5 ±0.5 ±3.0 ±3.0 4.5/4.5 5.4/3.6 ±5.0

4.2 Performance indices

The performance of chiller sequencing control strategies was evalauted by three


indices, viz. the total chiller switch number (Nt), the energy use of the chiller plant (En),
and the under-cooling percentage (Puc). The total switch number was used to quantify
the switch frequency, which was increased by 1 whenever one chiller was switched on
or off. The energy use was defined to indicate the energy performance of chiller plant,
which is the sum of energy consumption of all the chillers and pumps. The
under-cooling percentage was defined to indicate insufficient cooling provided by the
chiller plant. If the chiller plant saves energy in the price of insufficient cooling, the
energy saving is meaningless. This study assumed that the cooling supply was
insufficient when the CHWS temperature was higher than its setpoint by a certain
value, say ε. Therefore, the under-cooling percentage was calculated by

1 1, if Tsup,act ,k > Tsup,r + ε


=
PUC ∑
τ k
ζ k ∆t with ζ k =  (6)
0, otherwise

To evaluate the robustness of the hybrid control, the performance should be compared
with a benchmark case (the Q-based control without considering uncertainties).
Therefore, the variations of the performance indices, viz. the total switch number
percentage variation ∆𝑁𝑡 , the increment of the under-cooling percentage ∆𝑃𝑈𝐶 and the
energy use percentage variation ∆𝐸𝑛 , were defined as

N t − N t ,0 En − En , 0
∆N t = × 100% , ∆PUC = PUC − PUC ,0 , ∆En = × 100% (7)
N t ,0 En , 0
where 𝑁𝑡,0 ,𝑃𝑈𝐶,0,𝐸𝑛,0are the total switch number, the under-cooling percentage and the
energy use of the benchmark case, respectively.

100
The 7th Greater Pearl River Delta Conference on Building Operation and Maintenance

4.3 Results and analysis

4.3.1 Robustness analysis of the hybrid control


Figure 3 shows the robustness analysis results of the hybrid control when it suffered
from individual uncertainties. It can be seen that when the measurement uncertainties
increased from Level 1 to Level 5, the performance indices of the hybrid control were
changed insignificantly. The energy consumption kept almost the same as the
benchmark with the maximum variation being 0.06%. The total switch number was
increased from 0.19% to 0.37% at Levels 1-2 and then was decreased to -0.74% at
Level 4 and -0.37% at Level 5. The under cooling percentage was increased from
0.05% to 0.31% with the increase of the uncertainty levels.

When the control uncertainty was considered, the variation of the total switch number
did not change at Levels 1-3 and then was increased from 0.19% to 0.74% at Levels
4-5. The energy consumption was decreased to -0.03% at Level 1 but was increased
monotonously from 0.03% to 0.15% from Level 2 to Level 4. The under cooling
percentage was increased to 0.05% at Level 1 but was decreased to -0.51%
monotonously against the increase of uncertainty levels. Although the hybrid control
shows irregular variations to the control uncertainty, the maximum variations of
performance indices were less than ±1%.

When the operational uncertainty was considered, the total switch number was kept the
same as benchmark and the energy consumption was only increased to 0.19% with the
increase of uncertainty levels. The under cooling percentage was increased from about
0.06% to 0.18% at Levels 1-2 but was decreased to -0.37% at Level 5. Such variations
less than ±1% indicate that the hybrid control is robust to operational uncertainty.

When the hybrid control suffered from the threshold uncertainty, the energy
consumption was increased from 0% to 0.11% monotonously while the under cooling
percentage decreased to 0.2% as the uncertainty levels increased. The total switch
number showed irregular variations to the uncertainty increase with a maximum
variation to be 0.74% at Level 1. Again, the maximum variations of the three
performance indices were smaller than ±1%.

In comparison with the 10% performance degradation as defined for good robustness
in Liao et al. [6], the hybrid control has much better robustness since all the
performance variations were within ±1%. Therefore, it can be concluded that proposed
hybrid control is robust to all the individual uncertainties it suffers from.

101
The 7th Greater Pearl River Delta Conference on Building Operation and Maintenance

1.0 1.0
Performance variation (%)

Performance variation (%)


∆En ∆Nt ∆PUC ∆En ∆Nt ∆PUC
0.5 0.5

0.0 0.0

-0.5 -0.5

-1.0 -1.0
Level 1 Level 2 Level 3 Level 4 Level 5 Level 1 Level 2 Level 3 Level 4 Level 5
(a) Levels of measurement uncertainty (b) Levels of control uncertainty
1.0 1.0
Performance variation (%)

Performance variation (%)


∆En ∆Nt ∆PUC ∆En ∆Nt ∆PUC
0.5 0.5

0.0 0.0

-0.5 -0.5

-1.0 -1.0
Level 1 Level 2 Level 3 Level 4 Level 5 Level 1 Level 2 Level 3 Level 4 Level 5
(c) Levels of operational uncertainty (d) Levels of threshold uncertainty

Figure 3 - Performance variations of hybrid control when subjecting to different


levels of: a) measurement uncertainties; b) control uncertainties; c) operational
uncertainties; d) threshold uncertainties

4.3.2 Performance comparison with conventional controls


Figure 4 compares the performance of the hybrid control with the conventional T-based
control and P-based control when suffering from combined uncertainties at the same
level. It can be seen that when compared to the performance variations of the
conventional T-based control and P-based control, the performance variations of the
hybrid control were much smaller. When regarding to the energy consumption, the
hybrid control and the T-based control had similar variations with maximum increase
as 0.4% but the variation of the P-based control was up to -4.8%. When regarding to
the total switch number, the one of the hybrid control had a maximum variation as
1.1% at Level 4 while the one of the T-based control was increased to 30.2% at Level 5
and the one of the P-based control was decreased from -4.5% to -15.6% from Level 1
to Level 5. When regarding to the under cooling percentage, the variation of the hybrid
control showed a decrease from 0.1% to -0.6% against the increase of uncertainty
levels while the variation of the T-based control was increased from -0.9% to 3.3% and
the variation of the P-based control was increased from -0.1% to 12.1%.

The significantly reduced variations of all performance indices demonstrate that the
hybrid control has much stable performance under uncertainties and has much better
robustness than the conventional controls. The stable performance and good robustness
will enable it to be more reliable for applications under the uncertain operating
conditions.

102
The 7th Greater Pearl River Delta Conference on Building Operation and Maintenance

1.0 40.0 14.0

Variation of switch number (%)


Variation of energy use (%)
Hybrid control 12.0 Hybrid control

Variation of under cooling


0.0 30.0 T-based control
-1.0 T-based control 10.0

percentage (%)
20.0 P-based control
P-based control 8.0
-2.0
10.0 6.0
-3.0
Hybrid control 4.0
0.0
-4.0
T-based control 2.0
-5.0 -10.0
P-based control 0.0
-6.0 -20.0 -2.0
Level 1 Level 2 Level 3 Level 4 Level 5 Level 1 Level 2 Level 3 Level 4 Level 5 Level 1 Level 2 Level 3 Level 4 Level 5
(a) Levels of uncertainties (b) Levels of uncertainties (c) Levels of uncertainties

Figure 4 - Performance comparison between the hybrid control, the T-based


control and P-based control: a) energy consumption; b) total switch number; c)
under cooling percentage

5 Conclusions

This paper presents a hybrid control strategy for chiller sequencing control with
enhanced robustness to uncertainties. The proposed control hybridized two typical
controls to determine the switching actions. In the hybrid control, the chilled water
return temperature-based control was used to determine the switch-on action and the
direct power-based control was used to determine the switch-off action. The power
condition of online chillers was simultaneously used to correct the switch-on actions
for further reducing the unnecessary switch-on actions. The performance of the hybrid
control was evaluated by robustness analysis subjecting to multiple levels of
uncertainties and comparing with conventional controls. The robustness analysis
demonstrated that the developed hybrid control was robustness to all the uncertainties
it suffers from. The comparison studies showed that the hybrid control had
significantly improved performance for chiller sequencing control. Such good
robustness and improved control performance indicates that the developed hybrid
control is proper for real applications as multiple uncertainties exist.

6 Acknowledgements

The work described in this paper was fully supported by a grant from the Research
Grants Council of the Hong Kong Special Administrative Region, China (No.
11201215).

7 References

1. Huang G.S., Wang S.W., Sun Y.J. ‘Enhancing the Reliability of Chiller Control
Using Fused Measurement of Building Cooling Load’, HVAC&R Research.
Volume14, 2008.
2. Chang Y.C., Lin F.A., Lin C.H. ‘Optimal chiller sequencing by branch and bound
method for saving energy’, Energy Conversion and Management. Volume 46,
2005.

103
The 7th Greater Pearl River Delta Conference on Building Operation and Maintenance

3. Honeywell. ‘Engineering manual of automatic control for commercial buildings’,


Minneapolis, MN: Honeywell SI Edition, 1997.
4. Kwan C. ‘Investigation on sensing topology for sequential control of chiller plants
for efficient operation and control’, Master thesis, Hong Kong Polytechnic
University, 2001.
5. Huang G.S., Sun Y.J., Wang S.W. ‘Building instantaneous cooling load fused
measurement: multiple-sensor-based fusion versus chiller-model-based fusion’,
Building Services Engineering Research & Technology, Volume 34, Number 2,
2013.
6. Liao Y.D., Huang G.S., Sun Y.J., Zhang L.F. ‘Uncertainty analysis for chiller
sequencing control’, Energy and Buildings, Volume 85, 2014.
7. Liao Y.D., Sun Y.J., Huang G.S. ‘Robustness analysis of chiller sequencing
control’, Energy Conversion and Management, Volume 103, 2015.
8. Jiang W., Reddy T.A., Gurian P. ‘General methodology combining engineering
optimization of primary HVAC&R plants with decision analysis methods-Part II:
Uncertainty and decision analysis’, HVAC&R Research, Volume 13, 2007.
9. Huang G.S., Wang S.W., Xiao F., Sun Y.J. ‘A data fusion scheme for building
automation systems of building central chilling plants’, Automation in Construction,
Volume 18, Number 3, 2009.
10. Li Z.W., Huang G.S., Sun Y.J. ‘Stochastic chiller sequencing control’, Energy and
Buildings, Volume 84, 2014.
11. Wang S.W. ‘Intelligent buildings and building automation’, London & New York:
Spon Press, 2010.
12. JCGM100, ‘Evaluation of measurement data - Guide to the expression of
uncertainty in measurement’, 2008.
13. TRNSYS, ‘TRNSYS 17 Documentation’, 2012.
14. Schwedler M., Yates A. ‘Multiple-chiller-system design and control. Applications
Engineering Manual (SYS-APM001-EN)’, TRANE, 2001.
15. EDR Technical Report, ‘Chilled water plant design guideline’, 2009.

104

S-ar putea să vă placă și