Professional Documents
Culture Documents
∗ Corresponding author.
E-mail addresses: jiayao_neu@163.com (Y. Jia), tychai@mail.neu.edu.cn (T. Chai), mikewanguk@yahoo.co.uk (H. Wang), cysu@alcor.concordia.ca
(C.-Y. Su).
https://doi.org/10.1016/j.conengprac.2020.104372
Received 29 December 2019; Received in revised form 7 March 2020; Accepted 7 March 2020
Available online 20 March 2020
0967-0661/© 2020 Elsevier Ltd. All rights reserved.
Y. Jia, T. Chai, H. Wang et al. Control Engineering Practice 98 (2020) 104372
2
Y. Jia, T. Chai, H. Wang et al. Control Engineering Practice 98 (2020) 104372
[ ]
𝐶1 (⋅) 𝑇1 − 𝑇3 (𝑡) where,
𝐾12 (⋅) = ,
𝑀2 𝐶2
𝐴2 (𝑧−1 ) = 1 + 𝑎21 𝑧−1 ,
𝜌2 𝐶2 𝐹2 (⋅)𝑇2 (⋅) + 𝐴𝑈 (⋅)𝑇1 − 𝐴𝑈 (⋅)𝑇2 (⋅) 𝐵2 (𝑧−1 ) = 𝑏20 .
𝑑1 (⋅) =
𝑀2 𝐶2
𝑑𝑇 (𝑡) . In the above equations, 𝑣1 (𝑇 ) and 𝑣2 (𝑘) represent the dynamic
𝐴𝑈 (⋅)𝑇3 (𝑡) − 𝑀1 (⋅)𝐶1 (⋅) 𝑑𝑡3
+ changes in Eqs. (1) and (2) of the industrial HES and the modeling
𝑀2 𝐶2 errors in Eqs. (5) and (6). It has been validated from experiments that
𝑑𝑦2 (𝑡) 𝑣1 (𝑇 ) and 𝑣2 (𝑘) are bounded.
= 𝐾21 𝑦2 (𝑡) + 𝐾22 (⋅)𝑢2 (𝑡), (2)
𝑑𝑡
2.3. The control objectives
where,
1 The considered industrial HES is in an environment with large am-
𝐾21 = − ,
𝜏2 bient temperature variations. In order to ensure the targeted delivery
√ temperature of the hot water to users at its specified value, the set-point
𝑘2 (⋅) 𝛥𝑃2 (⋅)
𝐾22 (⋅) = . of the SWT should be adjusted according to the changes of ambient
𝜏2 temperature, and the tracking error between the SWT and set-point
The physical meanings of variables and parameters in Eqs. (1) and temperature is required to be controlled within the range specified
(2) are shown in Table 1. by the process. Additionally, in order to ensure the safe operation
It has been observed that parameters 𝐾11 (⋅), 𝐾12 (⋅) and 𝐾22 (⋅) in of the HES, it is necessary to minimize the fluctuation of the steam
Eqs. (1) and (2) in the model of residential HES established in Bastida valve. Therefore, the control objective is to design an adaptive cascade
et al. (2017) are constants, and 𝑑1 (⋅) in (1) is measurable interference. controller for the plant to be controlled given in (5) and (6), with the
However, as mentioned in the introduction section, the circulating steam regulating valve signal 𝑢2 (𝑘) as the control input, the SFR 𝑦2 (𝑘)
water temperature 𝑇2 (⋅) of the investigated industrial HES in this paper as the inner-loop output and the SWT 𝑦1 (𝑇 ) as the outer-loop output.
generally exhibits large variations as the ambient temperature changes When the system is subjected to a wide range of unknown disturbances
significantly during the system operation. The heat transfer coefficient and model uncertainties, it is expected that the tracking error 𝑒1 (𝑇 )
𝑈 (⋅) is also subjected to unknown changes due to precipitates left on the between the SWT 𝑦1 (𝑇 ) and the set-point 𝑦1𝑠𝑝 (𝑇 ) can be controlled
inner wall of the plate heat exchanger, which are produced by Ca2+ and within the targeted range under all operating conditions. Therefore, one
Mg2+ minerals in the water. In addition, the steam used in the process control objective is expressed as follows:
of intermittent production of desalted water, starch, and collector, etc.
| |
will cause the steam pressure 𝑃1 (⋅) to fluctuate within the range of |𝑦1𝑠𝑝 (𝑇 ) − 𝑦1 (𝑇 )| = ||𝑒1 (𝑇 )|| ≤ 𝛿1 1 ≤ 𝑇 < ∞, (7)
| |
0.25 MPa – 0.75 MPa. Therefore, unlike the original model in Bastida Furthermore, fluctuations of SFR 𝑦2 (𝑘) should be controlled within the
et al. (2017), the parameters {𝐾11 (⋅), 𝐾12 (⋅), 𝐾22 (⋅)} and the interference target range, namely
term 𝑑1 (⋅) of the model equations (1) and (2) are also time-varying and
unknown in this paper. |𝑦2 (𝑘) − 𝑦2 (𝑘 − 1)| = |𝛥𝑦2 (𝑘)| ≤ 𝛿2 1 ≤ 𝑘 < ∞, (8)
| | | |
For simplicity but without loss of generality, 𝐾11 (⋅), 𝐾12 (⋅) and 𝐾22 (⋅)
where 𝛿1 is the upper bound of the SWT tracking error 𝑒1 (𝑇 ), 𝛿2 is
are replaced by unknown constants 𝐾 11 , 𝐾 12 and 𝐾 22 , respectively.
the upper bound of the SFR change rate 𝛥𝑦2 (𝑘). 𝛿1 and 𝛿2 are preset
Together with an unknown function, the Eqs. (1) and (2) can be further
according to the process requirements.
expressed as follows.
𝑑𝑦1 (𝑡) 3. The proposed control method
= 𝐾 11 𝑦1 (𝑡) + 𝐾 12 𝑦2 (𝑡) + 𝑣1 (⋅), (3)
𝑑𝑡
3.1. Control strategy
𝑑𝑦2 (𝑡)
= 𝐾21 𝑦2 (𝑡) + 𝐾 22 𝑢2 (𝑡) + 𝑣2 (⋅), (4)
𝑑𝑡 In Rolf Isermann (1981), the discrete Z-N tuning method is used
where, to design the cascade PI control system of an industrial heat exchange
process. Frequent unknown changes in steam pressure, outdoor tem-
𝑣1 (⋅) = 𝐾11 (⋅)𝑦1 (𝑡) − 𝐾 11 𝑦1 (𝑡) + 𝐾12 (⋅)𝑦2 (𝑡) − 𝐾 12 𝑦2 (𝑡) + 𝑑1 (⋅), perature and circulating water quality will usually cause unknown
𝑣2 (⋅) = 𝐾22 (⋅)𝑢2 (𝑡) − 𝐾 22 𝑢2 (𝑡). variations of 𝑣1 (𝑇 ) and 𝑣2 (𝑘) in (5) and (6). In this case, it becomes
difficult for the integrator (1 − 𝑧−1 ) in the cascade PI to eliminate the
Let T be the sampling time instant of SWT, with 𝑇 = 1 representing influence of 𝑣1 (𝑇 ) and 𝑣2 (𝑘) on SWT and SFR, and thus increases the
a sampling period of 5 s; and denote k as the sampling instant of SFR, tracking error fluctuation. It can be seen from (3) and (4) that un-
with 𝑘 = 1 representing a sampling period of 1 s. The Euler method modeled dynamic 𝑣1 (𝑇 ) and 𝑣2 (𝑘) contain unknown nonlinear function
is then used to discretize (3) and (4). By using the actual data of SWT terms, so it is difficult to adopt self-tuning control method based on the
𝑦1 (𝑇 ), SFR 𝑦2 (𝑘) and valve opening 𝑢2 (𝑘), the well-known least-squares parameter estimation, such as the work by Song, Cai, and Wang (2003)
method can be applied to estimate the linear model parameters, and and anti-interference control method based on disturbance observer
the discretized linear model for the system model given in (3) and (4) (e.g., the work by Wu, Zhang, & Sun, 2018). To tackle the aforemen-
can then be expressed as follows: tioned challenges, in this paper a novel signal compensation method is
𝐴1 (𝑧−1 )𝑦1 (𝑇 + 1) = 𝐵1 (𝑧−1 )𝑦2 (𝑇 ) + 𝑣1 (𝑇 ), (5) used to improve the performance of the cascade PI control system.
Using Eqs. (5) and (6), the compensation signals can be designed
where, to eliminate the influences of 𝑣1 (𝑇 ) and 𝑣2 (𝑘). 𝑣1 (𝑇 ) and 𝑣2 (𝑘) can be
expressed by the values of 𝑣1 (𝑇 − 1) and 𝑣2 (𝑘 − 1) at the previous
𝐴1 (𝑧−1 ) = 1 + 𝑎11 𝑧−1 ,
sampling instant and the change rates of 𝛥𝑣1 (𝑇 ) and 𝛥𝑣2 (𝑘). For nota-
𝐵1 (𝑧−1 ) = 𝑏10 . tional simplicity, 𝑣(𝑘) is used herein to express 𝑣1 (𝑇 ) and 𝑣2 (𝑘). Because
𝑣(𝑘−1) can be obtained accurately, a compensation signal is designed to
and,
eliminate the influence of 𝑣(𝑘 − 1) on the controlled variable by using a
𝐴2 (𝑧−1 )𝑦2 (𝑘 + 1) = 𝐵2 (𝑧−1 )𝑢2 (𝑘) + 𝑣2 (𝑘), (6) feedforward control law. Although 𝛥𝑣(𝑘) is not measurable, the tracking
3
Y. Jia, T. Chai, H. Wang et al. Control Engineering Practice 98 (2020) 104372
Table 1
Variables and parameters of the concerned industrial heat exchange process.
Model variables Description Disturbance variables Description Model parameters Description
𝑦1 (t ) The SWT 𝑃1 (⋅) Steam pressure 𝐾11 (⋅), 𝐾12 (⋅) Model parameters for the SWT
𝑦2 (t ) The SFR 𝑃2 (⋅) Differential pressure of steam regulating valve 𝐾21 , 𝐾22 (⋅) Model parameters for the SFR
𝑢2 (t ) The steam valve opening 𝑇1 Steam temperature A Heat transfer area
𝑇2 (⋅) Circulating water temperature U (⋅) Heat transfer coefficient
𝑇3 (t ) Condensate temperature 𝑀2 Quality of circulating water
𝐹2 (⋅) Circulating water flow rate 𝐶2 Specific heat of circulating water
𝜌2 Circulating water density
𝑀1 (⋅) Steam quality
𝐶1 (⋅) Steam specific heat
𝑘2 (⋅) SFR coefficient
𝜏2 Regulating valve time constant
0.9𝑇𝐺2 0.135𝑇𝐺2
𝑘𝑝2 = −
𝑇𝑈 2 + 1∕2 (𝑇𝑈 2 + 1∕2)2
. (12)
0.27𝑇𝐺2
𝑘𝑖2 =
Fig. 3. The control structure of the signal compensator. (𝑇𝑈 2 + 1∕2)2
(2) Design of the signal compensator
As we discussed before, 𝑣2 (𝑘) in Eq. (6) can be expressed as:
error 𝑒(𝑘) caused by SWT and SFR can be measured, and 𝛥𝑣(𝑘 − 1) can
then be accurately obtained. Therefore, 𝛥𝑣(𝑘 − 1) and 𝑒(𝑘) are used to 𝑣2 (𝑘) = 𝑣2 (𝑘 − 1) + 𝛥𝑣2 (𝑘). (13)
design compensation signals to eliminate the influence of 𝛥𝑣(𝑘) on the According to Eq. (6), 𝑣2 (𝑘 − 1) is calculatable from:
tracking error 𝑒(𝑘), and the compensation signal 𝑢2 (𝑘) is superimposed
on the output 𝑢1 (𝑘) of the existing PI controller. The proposed PI 𝑣2 (𝑘 − 1) = 𝐴2 (𝑧−1 )𝑦2 (𝑘) − 𝐵2 (𝑧−1 )𝑢2 (𝑘 − 1). (14)
controller based on signal compensation method is therefore shown in
According to Eq. (13), 𝛥𝑣2 (𝑘 − 1) can be expressed as:
Figs. 2 and 3.
In the classical cascade control, the closed-loop system for the inner- 𝛥𝑣2 (𝑘 − 1) = 𝑣2 (𝑘 − 1) − 𝑣2 (𝑘 − 2). (15)
loop is required to reach steady-state during each sampling period of
the outer-loop control system, while in reality the closed-loop system Substituting (10) and (11) into (9), 𝑢2 (k) is obtained as follows:
for the inner-loop of SFR may have dynamic changes during the sam- [ ]
pling of the outer-loop control of SWT. For this reason, a dynamic 𝑢2 (𝑘) = 𝑢2 (𝑘 − 1) + 𝑘𝑝2 𝑒2 (𝑘) − 𝑒2 (𝑘 − 1) + 𝑘𝑖2 𝑒2 (𝑘)
. (16)
model of the outer-loop of SWT reflecting the characteristics of the +(1 − 𝑧−1 )𝑢22 (𝑘) + (1 − 𝑧−1 )𝑢23 (𝑘)
closed-loop system for the inner-loop of SFR is established, and the
model is designed as shown in Figs. 2 and 3. A cascade PI control system Substituting Eq. (16) into Eq. (6) of plant to be controlled of SFR,
based on the signal compensation method is thus proposed as shown in it can be further obtained that:
Fig. 4 [ ]
𝐴2 (𝑧−1 )(1 − 𝑧−1 ) + 𝑧−1 𝑏20 (𝑘𝑝2 + 𝑘𝑖2 − 𝑘𝑝2 𝑧−1 ) 𝑒2 (𝑘 + 1)
3.2. Cascade PI control algorithm based on the signal compensation method = 𝐴2 (𝑧−1 )(1 − 𝑧−1 )𝑦2𝑠𝑝 (𝑘 + 1)
[ ] . (17)
In the signal compensation based cascade PI control system as −(1 − 𝑧−1 ) 𝑏20 𝑢22 (𝑘) + 𝑣2 (𝑘 − 1)
shown in Figs. 2–4, firstly, the inner-loop PI controller of SFR with −(1 − 𝑧−1 )𝑏20 𝑢23 (𝑘) − (1 − 𝑧−1 )𝛥𝑣2 (𝑘)
4
Y. Jia, T. Chai, H. Wang et al. Control Engineering Practice 98 (2020) 104372
It can be seen from Eq. (17) that the effect of 𝑣2 (𝑘 − 1) on tracking The output 𝑢2 (𝑘) of PI controller in the inner-loop of SFR based on
error 𝑒2 (𝑘) can be eliminated by making 𝑏20 𝑢22 (𝑘) + 𝑣2 (𝑘 − 1) = 0, so the signal compensation is obtained by Eq. (9), wherein, the output
that the feedforward compensation signal for 𝑣2 (𝑘 − 1) can be readily 𝑢21 (𝑘) of PI controller is given by Eq. (11), PI parameters 𝑘𝑝2 and 𝑘𝑖2
obtained to give: are obtained by Eq. (12), and the output 𝑢′21 (𝑘) of signal compensator
1 is obtained by Eqs. (10), (18) and (27).
𝑢22 (𝑘) = − 𝑣 (𝑘 − 1). (18)
𝑏20 2
Thus, from (17) and (18), it can be obtained that: 3.2.2. The dynamic model of SWT based on inner-loop control system of
SFR
[ ]
𝐴2 (𝑧−1 )(1 − 𝑧−1 ) + 𝑧−1 𝑏20 (𝑘𝑝2 + 𝑘𝑖2 − 𝑘𝑝2 𝑧−1 ) 𝑒2 (𝑘 + 1) Because of the unknown disturbance and the unknown change of
. (19) model parameters, the closed-loop system for the SFR inner-loop is in
= −(1 − 𝑧−1 )𝑏20 𝑢23 (𝑘) − (1 − 𝑧−1 )𝛥𝑣2 (𝑘)
dynamical variations during the sampling period of the SWT outer-loop.
+𝐴2 (𝑧−1 )(1 − 𝑧−1 )𝑦2𝑠𝑝 (𝑘 + 1) Therefore, when designing the SWT outer-loop controller, the dynamic
It can be seen from (19) that 𝑒2 (𝑘 + 1) represents the influence of characteristics of the closed-loop system for the SFR inner-loop must
𝛥𝑣2 (𝑘) on the SFR 𝑦2 (𝑘). In order to eliminate the influence of 𝛥𝑣2 (𝑘) be considered. Therefore, the closed-loop system for the SFR inner-loop
on 𝑦2 (𝑘), a compensator 𝑢23 (𝑘) is designed to minimize 𝑒2 (𝑘+1). For this can be obtained from (19), (25) and (27) to give:
purpose, the performance index of one-step optimal rhythm regulation 𝑦2 (𝑘 + 1) = 𝑦2𝑠𝑝 (𝑘 + 1) + 𝛥𝑣2 (𝑘). (28)
in Goodwin and Sin (1984) is adopted:
[ ]2 In order to establish the dynamical model of the SWT outer-loop
𝐽2′ = 𝑒2 (𝑘 + 1) . (20)
reflecting the dynamic characteristics of the closed-loop system for the
Let the partial derivative of 𝐽2′ with respect to 𝑢23 (𝑘) in (20) be zero, inner-loop of SFR, the closed-loop system given in (28) at 𝑘 time is first
that is: transformed into the closed-loop equation at 𝑇 time of the sampling
𝜕𝐽2′ period of the outer-loop control of SWT. Since the set-point 𝑦2𝑠𝑝 (𝑘)
𝜕𝑒2 (𝑘 + 1)
= 2𝑒2 (𝑘 + 1) = 0. (21) generated by the outer-loop control is constant within the sampling
𝜕𝑢23 (𝑘) 𝜕𝑢23 (𝑘)
period 𝑇 , 𝑇 = 5𝑘, the following equality is true:
From Eq. (19), it can be seen that:
𝑦2𝑠𝑝 (5𝑘) = 𝑦2𝑠𝑝 (5𝑘 + 𝑖) (𝑖 = 1, 2, 3, 4). (29)
𝜕𝑒2 (𝑘 + 1)
= −𝑏20 . (22)
𝜕𝑢23 (𝑘) According to Eqs. (28) and (29), the closed-loop system equation of
By substituting Eq. (22) into Eq. (21), one can obtain the regulation SFR with respect to the time of sampling cycle T of SWT outer-loop can
law of minimizing 𝐽2′ as follows: be obtained as follows:
In order to find the one-step optimal prediction 𝑒∗2 (𝑘 + 1∕𝑘) of where 𝑣̃2 (𝑇 ) = 𝑣2 (5𝑘 + 4).
𝑒2 (𝑘 + 1), the following Diophantine equation is introduced: Eq. (30) represents the SFR system with respect to the time of sam-
pling period controlled by the outer-loop of SWT, and the dynamical
𝐴2 (𝑧−1 )(1 − 𝑧−1 ) + 𝑏20 (𝑘𝑝2 + 𝑘𝑖2 − 𝑘𝑝2 𝑧−1 )𝑧−1 + 𝑧−1 𝐺2′ (𝑧−1 )
, (24) model of the outer-loop of SWT based on the closed-loop system of SFR
=1
can be obtained from the Eqs. (30) and (5) to give:
where 𝐺2′ (𝑧−1 ) is a polynomial about 𝑧−1 . According to Eq. (24):
𝐴1 (𝑧−1 )𝑦1 (𝑇 + 1) = 𝐵1 (𝑧−1 )𝑦2𝑠𝑝 (𝑇 ) + 𝑣(𝑇 ), (31)
𝐺2′ (𝑧−1 ) −1
= 1 − 𝑎21 − 𝑏20 𝑘𝑝2 − 𝑏20 𝑘𝑖2 + (𝑎21 + 𝑏20 𝑘𝑝2 )𝑧 . (25)
where 𝑣(𝑇 ) = 𝑣1 (𝑇 ) + 𝐵1 (𝑧−1 )̃
𝑣2 (𝑇 − 1).
Substituting Eq. (24) into Eq. (19), we can obtain:
3.2.3. The outer-loop PI controller of SWT based on signal compensation
𝑒2 (𝑘 + 1) = 𝐺2′ (𝑧−1 )𝑒2 (𝑘) − 𝑏20 (1 − 𝑧−1 )𝑢23 (𝑘) The PI controller of SWT based on the signal compensation method
. (26)
+𝐴2 (𝑧−1 )(1 − 𝑧−1 )𝑦2𝑠𝑝 (𝑘 + 1) + 𝛥𝑣2 (𝑘 − 1) − 𝛥𝑣2 (𝑘) is as follows:
According to Eqs. (23) and (26), the compensation signal 𝑢23 (𝑘) can 𝑦2𝑠𝑝 (𝑇 ) = 𝑦2𝑠𝑝1 (𝑇 ) + 𝑦′2𝑠𝑝1 (𝑇 ), (32)
be readily obtained as follows:
where 𝑦2𝑠𝑝1 (𝑇 ) is the output of PI controller for the outer-loop and
𝐺2′ (𝑧−1 )
𝑢23 (𝑘) = 𝑢23 (𝑘 − 1) + 𝑒2 (𝑘) 𝑦′2𝑠𝑝1 (𝑇 ) is the output of signal compensator, which can be further
𝑏20 expressed as:
. (27)
𝐴2 (𝑧−1 )(1 − 𝑧−1 ) 1
+ 𝑦2𝑠𝑝 (𝑘 + 1) + 𝛥𝑣 (𝑘 − 1) 𝑦′2𝑠𝑝1 (𝑇 ) = 𝑦2𝑠𝑝2 (𝑇 ) + 𝑦2𝑠𝑝3 (𝑇 ), (33)
𝑏20 𝑏20 2
5
Y. Jia, T. Chai, H. Wang et al. Control Engineering Practice 98 (2020) 104372
where 𝑦2𝑠𝑝2 (𝑇 ) is the feedforward compensation signal of 𝑣(𝑇 − 1) and Using the design method of dynamic rate of change compensator
𝑦2𝑠𝑝3 (𝑇 ) is the compensation signal of 𝛥𝑣(𝑇 ). In the following, detailed without modeling the SFR, the compensation signal 𝑦2𝑠𝑝3 (𝑇 ) which can
design procedure will be given. minimize the performance index 𝐽1′ can be obtained as follows:
(1) PI controller design
𝑏10 𝐺1′ (𝑧−1 )
The incremental PI controller for the outer-loop is again in the 𝑦2𝑠𝑝3 (𝑇 ) = 𝑦2𝑠𝑝3 (𝑇 − 1) + 𝑒1 (𝑇 )
following form: 𝜆1 2 + 𝑏10 2
𝑏10 𝐴1 (𝑧−1 )(1 − 𝑧−1 )
[ ] + 𝑦1𝑠𝑝 (𝑇 + 1) , (42)
𝑦2𝑠𝑝1 (𝑇 ) = 𝑦2𝑠𝑝1 (𝑇 − 1) + 𝑘𝑝1 𝑒1 (𝑇 ) − 𝑒1 (𝑇 − 1) + 𝑘𝑖1 𝑒1 (𝑇 ), (34) 𝜆1 2 + 𝑏10 2
𝑏10
where 𝑒1 (𝑇 ) = 𝑦1𝑠𝑝 (𝑇 ) − 𝑦1 (𝑇 ), 𝑒1 (𝑇 ) is the tracking error of SWT, 𝑘𝑝1 + 𝛥𝑣(𝑇 − 1)
and 𝑘𝑖1 are the proportional gain and integral gain of PI controller of 𝜆1 2 + 𝑏10 2
SWT, respectively. where 𝐺1′ (𝑧−1 ) is a polynomial of 𝑧−1 , and is expressed as:
The discrete Z-N tuning method as given in Rolf Isermann (1981) is
again used to select 𝑘𝑝1 and 𝑘𝑖1 . First, PI controllers given in (11) and 𝐺1′ (𝑧−1 ) = 1 − 𝑎11 − 𝑏10 𝑘𝑝1 − 𝑏10 𝑘𝑖1 + (𝑎11 + 𝑏10 𝑘𝑝1 )𝑧−1 . (43)
(12) are adopted as the SFR inner-loop PI control. Then a set of step In order to choose the weighting coefficient 𝜆1, substituting (32)–
response experiments are carried out on the outer-loop of SWT with the (34), (40) and (42) into the plant to be controlled given in (31), we can
SFR inner-loop PI control system. The delay time 𝑇𝑈 1 and the balance obtain the closed-loop equation with 𝑦2𝑠𝑝3 (𝑇 ) as the input and 𝑒1 (𝑇 + 1)
time 𝑇𝐺1 can be obtained. As a result, the parameters of PI controller as the output to give:
of SWT are given by:
⎧ ⎫
0.9𝑇𝐺1 0.675𝑇𝐺1 ⎪𝜆1 2 (1 − 𝑧−1 )𝐴1 (𝑧−1 ) ⎪
𝑘𝑝1 = − ⎨ [ ]⎬ 𝑒1 (𝑇 + 1)
𝑇𝑈 1 + 5∕2 (𝑇𝑈 1 + 5∕2)2 ⎪ +𝑏10 𝑏10 + 𝜆1 2 (𝑘𝑝1 + 𝑘𝑖1 )𝑧−1 − 𝜆1 2 𝑘𝑝1 𝑧−2 ⎪
. (35) ⎩ ⎭ . (44)
1.35𝑇𝐺1
𝑘𝑖1 = = 𝜆1 2 (1 − 𝑧−1 )𝐴1 (𝑧−1 )𝑦1𝑠𝑝 (𝑇 + 1)
(𝑇𝑈 1 + 5∕2)2 [ ]
− 𝜆1 2 (1 − 𝑧−1 ) + 𝑏10 2 𝛥𝑣(𝑇 )
(2) Design of signal compensator
For the outer-loop system, it can be seen that the unmodeled When the weighting coefficient 𝜆1 satisfies the following condition,
dynamics again satisfies the stability of the closed-loop system for the outer-loop SWT can be
ensured.
𝑣(𝑇 ) = 𝑣(𝑇 − 1) + 𝛥𝑣(𝑇 ). (36) 𝜆1 2 (1 − 𝑧−1 )𝐴1 (𝑧−1 )
[ ] . (45)
According to Eq. (31), 𝑣(𝑇 − 1) can be expressed as: +𝑏10 𝑏10 + 𝜆1 2 (𝑘𝑝1 + 𝑘𝑖1 )𝑧−1 − 𝜆1 2 𝑘𝑝1 𝑧−2 ≠ 0, |𝑧| > 1
Finally, the output 𝑦2𝑠𝑝 (T ) of the PI controller in the outer-loop
𝑣(𝑇 − 1) = 𝐴1 (𝑧−1 )𝑦1 (𝑇 ) − 𝐵1 (𝑧−1 )𝑦2𝑠𝑝 (𝑘 − 1). (37)
of SWT based on the signal compensation is obtained from (32)–(35),
Also, in line with Eq. (36), 𝛥𝑣(𝑇 − 1) can be represented as follows: (40), (42) and (45).
Substituting Eqs. (32)–(34) into the outer-loop model (31) of SWT, In this section, to verify the effectiveness of the proposed control
it can be obtained that: method, the comparison of the control performance of the method
[ ] given in this paper with that of other relevant control methods is
𝐴1 (𝑧−1 )(1 − 𝑧−1 ) + 𝑧−1 𝑏10 (𝑘𝑝1 + 𝑘𝑖1 − 𝑘𝑝1 𝑧−1 ) 𝑒1 (𝑇 + 1)
conducted. For this purpose, the cascade PI control method in Jeng
= 𝐴1 (𝑧−1 )(1 − 𝑧−1 )𝑦1𝑠𝑝 (𝑇 + 1) and Jyh-Cheng (2014) and the classical cascade PI control method are
[ ] . (39)
−(1 − 𝑧−1 ) 𝑏10 𝑦2𝑠𝑝2 (𝑇 ) + 𝑣(𝑇 − 1) chosen to compare with the proposed one. The simulation is carried
out on the same simulation model given in Jeng and Jyh-Cheng (2014).
−(1 − 𝑧−1 )𝑏10 𝑦2𝑠𝑝3 (𝑇 ) + (1 − 𝑧−1 )𝛥𝑣(𝑇 )
Given the following system transfer function,
Let 𝑏10 𝑦2𝑠𝑝2 (𝑇 ) + 𝑣(𝑇 − 1) = 0 to eliminate the influence of 𝑣(𝑇 − 1) 1
on tracking error 𝑒1 (𝑇 ), 𝑦2𝑠𝑝2 (𝑇 ) can be formulated as follows: 𝑌1 (𝑠) = 𝑌2 (𝑠)
(10𝑠 + 1)(4s + 1)(s + 1)2 , (46)
1 1
𝑦2𝑠𝑝2 (𝑇 ) = − 𝑣(𝑇 − 1). (40) + 𝐷1 (𝑠)
𝑏10 (10𝑠 + 1)(4s + 1)(s + 1)2
Eqs. (40) and (39) can be used with 𝑦2𝑠𝑝3 (𝑇 ) as input and tracking 𝑒−0.1𝑠 𝑒−0.1𝑠
error 𝑒1 (𝑇 + 1) as output. This gives the following dynamics 𝑌2 (𝑠) = 𝑈2 (𝑠) + 𝐷 (𝑠), (47)
1.9𝑠 + 1 1.9𝑠 + 1 2
[ ] where 𝑌1 (𝑠) is the output of the outer-loop, 𝑌2 (𝑠) is the output of the
𝐴1 (𝑧−1 )(1 − 𝑧−1 ) + 𝑧−1 𝑏10 (𝑘𝑝1 + 𝑘𝑖1 − 𝑘𝑝1 𝑧−1 ) 𝑒1 (𝑇 + 1)
inner-loop, and 𝑈2 (𝑠) is the control input. In the above equations,
= 𝐴1 (𝑧−1 )(1 − 𝑧−1 )𝑦1𝑠𝑝 (𝑇 + 1) . 𝑑1 (𝑡) and 𝑑2 (𝑡) are disturbances of the outer-loop and the inner-loop,
−(1 − 𝑧−1 )𝑏10 𝑦2𝑠𝑝3 (𝑇 ) + (1 − 𝑧−1 )𝛥𝑣(𝑇 ) respectively.
{
It can be seen from the above equation that 𝑒1 (𝑇 + 1) reflects the 0 0 ≤ 𝑡 < 280
𝑑1 (𝑡) = , (48)
1 else
influence of 𝛥𝑣(𝑇 ) on the SWT 𝑦1 (𝑇 ). In order to eliminate the influence
of 𝛥𝑣(𝑇 ) on 𝑦1 (𝑇 ), the compensator 𝑦2𝑠𝑝3 (𝑇 ) is designed. Moreover, in {
order to make 𝑒1 (T + 1) as small as possible and the fluctuation of 0 0 ≤ 𝑡 < 150
𝑑2 (𝑡) = . (49)
𝑦2𝑠𝑝3 (𝑇 ) is minimized, the performance index of the one-step optimal 10 else
control law in Clarke and Gawthrop (1975) is adopted as follows: Its set-point 𝑦1𝑠𝑝 (𝑡) in Jeng and Jyh-Cheng (2014) is generated from:
[ ]2 [ ]2
𝐽1′ = 𝑒1 (𝑇 + 1) + 𝜆1 (1 − 𝑧−1 )𝑦2𝑠𝑝3 (𝑇 ) , (41) {
0 𝑡 < 10
where 𝜆1 is the weighting coefficient. 𝑦1𝑠𝑝 (𝑡) = . (50)
1 10 ≤ 𝑡 ≤ 240
6
Y. Jia, T. Chai, H. Wang et al. Control Engineering Practice 98 (2020) 104372
+0.0093𝑦2𝑠𝑝3 (𝑇 − 3) + 1.4544𝑒1 (𝑇 )
1 ∑[
𝑁
]2
−0.9953𝑒1 (𝑇 − 1) + 1.1975𝑒1 (𝑇 − 2) MSE = 𝑦 (𝑘) − 𝑦(𝑘) , (57)
, (56) 𝑁 𝑘=1 𝑠𝑝
+0.0185𝑒1 (𝑇 − 3) + 2.4605𝑦1𝑠𝑝 (𝑇 + 1)
−3.568𝑦1𝑠𝑝 (𝑇 ) + 1.1818𝑦1𝑠𝑝 (𝑇 − 1) ∑
𝑁
| |
IAE = |𝑦𝑠𝑝 (𝑘) − 𝑦(𝑘)| , (58)
−0.0743𝑦1𝑠𝑝 (𝑇 − 2) + 2.4605𝛥𝑣(𝑇 − 1) | |
𝑘=1
where, where 𝑦𝑠𝑝 (𝑘) is the targeted value, 𝑦(𝑘) is the system output and N is
the number of experimental samples.
𝑣(𝑇 − 1) = 𝑦1 (𝑇 ) − 0.4501𝑦1 (𝑇 − 1) + 0.0302𝑦1 (𝑇 − 2)
From Figs. 5–7, Tables 2 and 3, it can be seen that the control effect
+0.0001𝑦1 (𝑇 − 3) − 0.3403𝑦2𝑠𝑝 (𝑘 − 1) .
of Jeng and Jyh-Cheng (2014) is better than that of classical cascade
−0.2361𝑦2𝑠𝑝 (𝑘 − 2) − 0.0038𝑦2𝑠𝑝 (𝑘 − 3)
PI control, but the control effect of our proposed method is better than
The controller proposed in this paper, the classical cascade PI and that of classical cascade PI control method and the method by Jeng
the one from Jeng and Jyh-Cheng (2014) are adopted in the simulation and Jyh-Cheng (2014), in which, the outer-loop control performance
7
Y. Jia, T. Chai, H. Wang et al. Control Engineering Practice 98 (2020) 104372
Fig. 7. The operational responses of the outer-loop output 𝑦1 , set-point 𝑦1𝑠𝑝 and the Fig. 8. The operational responses of the outer-loop output 𝑦1 , set-point 𝑦1𝑠𝑝 and the
inner-loop output 𝑦2 , set-point 𝑦2𝑠𝑝 , and control input 𝑢2 from this paper. inner-loop output 𝑦2 , set-point 𝑦2𝑠𝑝 , and control input 𝑢2 from the classical cascade PI
method.
Table 2
Performance evaluation results of the outer-loop output y 1 for the classical cascade PI
control method, the control method of Jeng and Jyh-Cheng (2014) and the proposed
control method.
𝑦1 IAE MSE
Classical cascade PI method 44.7529 0.205
The method in Jeng and Jyh-Cheng (2014) 43.5172 0.2273
The proposed control method 14.4686 0.11
Table 3
Performance evaluation results of the inner-loop output 𝑦2 for the classical cascade PI
control method, the control strategy of Jeng and Jyh-Cheng (2014) and the proposed
control method.
𝑦2 IAE MSE
Classical cascade PI method 66.3013 1.0328
The method in Jeng and Jyh-Cheng (2014) 47.0446 0.7789
The proposed control method 24.5833 0.6725
IAE and MSE are reduced by 67.67%, 46.34%, 66.75% and 51.61%,
respectively; the inner-loop control performance IAE and MSE are Fig. 9. The operational responses of the outer-loop output 𝑦1 , set-point 𝑦1𝑠𝑝 and the
reduced by 62.92%, 34.89%, 47.74% and 13.66%, respectively. inner-loop output 𝑦2 , set-point 𝑦2𝑠𝑝 , and control input 𝑢2 from Jeng and Jyh-Cheng
In order to compare the control effects of the above three methods (2014).
when the system is subjected to large and frequent disturbances, the
following testing signals with the simulation models given by Eqs. (46)
and (47) are used:
⎧0 𝑡 < 100
⎪ [ ]
𝑑1 (𝑡) = ⎨ (𝑡 − 15)𝜋 , (59)
⎪ 1.0sign sin else
⎩ 80
[ ]
(𝑡 − 10)𝜋
𝑑2 (𝑡) = 5.0sign sin . (60)
2
The simulation results of the three methods are shown in Figs. 8–10,
and the performance results of the three methods are given in Tables 4
and 5, respectively.
From Figs. 8–10, Tables 4 and 5, it can be seen that the control
effect of Jeng and Jyh-Cheng (2014) is better than that of classical
cascade PI control, but the control effect of our proposed method is
significantly better than that of classical cascade PI control method and
the one by Jeng and Jyh-Cheng (2014), in which the IAE and MSE
of outer-loop control performance are reduced significantly by 87.6%,
89.21%, 87.18% and 89.14% respectively; the IAE and MSE of inner-
Fig. 10. The operational responses of the outer-loop output 𝑦1 , set-point 𝑦1𝑠𝑝 and the
loop control performance are reduced by 74.27%, 67.8%, 68.82% and
inner-loop output 𝑦2 , set-point 𝑦2𝑠𝑝 , and control input 𝑢2 from this paper.
54.45%, respectively.
8
Y. Jia, T. Chai, H. Wang et al. Control Engineering Practice 98 (2020) 104372
Table 4
Performance evaluation results of the outer-loop output y 1 for the classical cascade PI
control method, the control strategy of Jeng and Jyh-Cheng (2014) and the proposed
control method.
𝑦1 IAE MSE
Classical cascade PI method 338.2389 3.5398
The method in Jeng and Jyh-Cheng (2014) 327.1064 3.5148
The proposed control method 41.9293 0.3818
Table 5
Performance evaluation results of the inner-loop output 𝑦2 for the classical cascade PI
control method, the control strategy of Jeng and Jyh-Cheng (2014) and the proposed
control method.
𝑦2 IAE MSE
Classical cascade PI method 780.3192 26.1308
The method in Jeng and Jyh-Cheng (2014) 643.9725 18.4719 Fig. 12. The control system hardware platform.
The proposed control method 200.759 8.4134
9
Y. Jia, T. Chai, H. Wang et al. Control Engineering Practice 98 (2020) 104372
Fig. 14. The responses of the SWT 𝑦1 and the SFR 𝑦2 , and the steam valve opening Fig. 16. The responses of the SWT 𝑦1 and the SFR 𝑦2 , and the steam valve opening
𝑢2 using the proposed method control. 𝑢2 using the classical cascade PI method.
rate both exceed the targeted value ranges by using the method in
Chai et al. (2017) and the classical cascade PI control method. It can
be seen from Fig. 15 that when the method in Chai et al. (2017) is
adopted, the fluctuation range of SWT is ±4.2 ◦ C, the fluctuation range
of SFR change rate 𝛥𝑦2 is ±0.65 t/h, which is 40% and 30% higher
than the targeted ranges. Fig. 16 shows that when adopting the classical
cascade PI control method, the fluctuation range of SWT is ±5.0 ◦ C,
and the fluctuation range of SFR 𝛥𝑦2 is ±0.8 t/h, which are 66.67%
and 60% higher than the targeted value ranges, respectively. As it
Fig. 15. The responses of the SWT 𝑦1 and the SFR 𝑦2 , and the steam valve opening
𝑢2 using the method in Chai et al. (2017).
can be seen from Fig. 14, when adopting the method in this paper,
the fluctuation range of SWT is ± 3 ◦ C, and the fluctuation range of
SFR change rate 𝛥𝑦2 is ±0.45 t/h. This indicates that the proposed
control strategy performs better and can achieve the control objectives
Using the identified model (63), the compensation signal of inner-
described in Eq. (61). Compared with the cascade PI control method,
loop PI controller is designed offline. The compensators of 𝑣2 (𝑘 − 1) and
𝛥𝑣2 (𝑘) are respectively of the following: the fluctuation of SWT and SFR is reduced by 40% and 43.75%,
respectively.
𝑢22 (𝑘) = −37.59𝑣2 (𝑘 − 1), (64)
6. Conclusions
𝑢23 (𝑘) = 𝑢23 (𝑘 − 1) + 60.72𝑒2 (𝑘) − 24.28𝑒2 (𝑘 − 1)
+37.59𝑦2𝑠𝑝 (𝑘 + 1) − 67.06𝑦2𝑠𝑝 (𝑘) . (65) In this paper, we proposed a cascaded PI control with signal com-
+29.47𝑦2𝑠𝑝 (𝑘 − 1) + 37.59𝛥𝑣2 (𝑘 − 1)
pensation method for an industrial HES. The proposed control strategy
The compensation signal of the outer-loop PI controller is designed consists of an inner-loop PI controller for the SFR based on the signal
using the identified model (62). The compensators of 𝑣(𝑇 −1) and 𝛥𝑣(𝑇 ) compensation and an outer-loop PI controller for the SWT again based
are respectively obtained as: on the signal compensation. The PI controller based on the signal
compensation method is composed of the PI controller designed by the
𝑦2𝑠𝑝2 (𝑇 ) = −15.7𝑣(𝑇 − 1), (66)
discrete Z-N tuning method and the former unmodeled dynamic and
its rate of change compensators. The results of industrial application
𝑦2𝑠𝑝3 (𝑇 ) = 𝑦2𝑠𝑝3 (𝑇 − 1) + 0.11𝑒1 (𝑇 ) − 0.05𝑒1 (𝑇 − 1) showed that when the heat exchange process is subjected to unknown
+0.06𝑦1𝑠𝑝 (𝑇 + 1) − 0.13𝑦1𝑠𝑝 (𝑇 ) + 0.06𝑦1𝑠𝑝 (𝑇 − 1). (67) changes of steam pressure, ambient temperature and circulating water
+0.06𝛥𝑣(𝑇 − 1) quality, the proposed method can control the tracking error of SWT
For the purpose of comparison, the control performance of the and fluctuation of SFR within the targeted ranges. Indeed, the method
proposed method is analyzed and compared to that of the classical proposed in this paper can change the control input of the cascade
cascade PI control method, and the method in Chai et al. (2017), using PI control system by accurately calculating the compensation signal,
the same PI control parameters (the difference only lies in the signal so that the cascade PI control system has abilities of suppressing the
compensator). In the method by Chai et al. (2017), the compensator interference of the control loops and realizing robust adaptability.
of the SWT outer-loop 𝛥𝑣(𝑇 ) adopts one-step optimal regulation rule
designed as follows:
Declaration of competing interest
𝑦2𝑠𝑝3 (𝑇 ) = 𝑦2𝑠𝑝3 (𝑇 − 1) + 26.68𝑒1 (𝑇 ) − 11.76𝑒1 (𝑇 − 1)
+15.7𝑦1𝑠𝑝 (𝑇 + 1) − 30.97𝑦1𝑠𝑝 (𝑇 ) . (68)
The authors declare that they have no known competing finan-
+15.27𝑦1𝑠𝑝 (𝑇 − 1) + 15.7𝛥𝑣(𝑇 − 1)
cial interests or personal relationships that could have appeared to
5.3. The analysis of control performance of industrial application influence the work reported in this paper.
10
Y. Jia, T. Chai, H. Wang et al. Control Engineering Practice 98 (2020) 104372
References Jeng, & Jyh-Cheng (2014). Simultaneous closed-loop tuning of cascade controllers based
directly on set-point step-response data. Journal of Process Control, 24(5), 652–662.
Bastida, H., Ugalde-Loo, C. E., Abeysekera, M., & Qadrdan, M. (2017). Dynamic Jia, Y., & Chai, T. (2017). A data-driven dual-rate control method for a heat exchanging
modeling and control of a plate heat exchanger. In 2017 IEEE conference on energy process. IEEE Transactions on Industrial Electronics, 64(5), 4158–4168.
internet and energy system integration (EI2). IEEE. Matko, D., & Skrjanc, I. (2000). Predictive functional control based on fuzzy model for
Bobal, V., Kubalcik, M., & Dostal, P. (2013). Identification and self-tuning predictive heat-exchanger pilot plant. IEEE Transactions on Fuzzy Systems, 8(6), 705–712.
control of heat exchanger. In International conference on process control (pp. Rolf Isermann (1981). Digital control systems. Berlin Heidelberg: Springer-Verlag.
219–224). IEEE. Shi, Zhaoyu (1999). The adjustment of heat supply system and control. Bei Jing: Tsinghua
Chai, T., Jia, Y., Wang, H., & Su, C. Y. (2017). Dual rate adaptive control for an indus- University press.
trial heat supply process using signal compensation approach. IFAC-PapersOnLine, Song, S., Cai, W., & Wang, Y. G. (2003). Auto-tuning of cascade control systems. ISA
50(1), 1877–1884. Transactions, 42(1), 63–72.
Clarke, D. W., & Gawthrop, P. J. (1975). Self-tuning controller. IEEE Proceedings Part Tridianto, E., Ariwibowo, T. H., Almasa, S. K., & Prasetya, H. E. G. (2017). Cascaded PID
D: Control Theory and Applications, 122(9), 929–934. temperature controller for FOPDT model of shell-and-tube heat exchanger based on
Fischer, M., Nelles, O., & Isermann, R. (1998). Adaptive predictive control of a heat Matlab/Simulink. In International electronics symposium on engineering technology and
exchanger based on a fuzzy model. Control Engineering Practice, 6(2), 259–269. applications (pp. 185–191). IEEE.
Fratczak, M., & Czubasiewicz, R. (2016). Cascade balance-based adaptive control of Wang, L., Jia, Y., Chai, T., & Xie, W. (2018). Dual rate adaptive control for mixed
heat supply system — Simulation validation. In International conference on methods separation thickening process using compensation signal based approach. IEEE
and models in automation and robotics (pp. 1170–1175). IEEE. Transactions on Industrial Electronics, 65(4), 3621–3632.
Ghanassi, M., Champagne, B., & Kabal, P. (2007). On the steady-state mean squared Wu, K., Zhang, Z., & Sun, C. (2018). Disturbance-observer-based output feedback
error of the fixed-point lms algorithm. Signal Processing, 87(12), 3226–3233. control of non-linear cascaded systems with external disturbance. IET Control Theory
Goodwin, G. C., & Sin, K. S. (1984). Adaptive filtering prediction and control. & Applications, 12(6), 738–744.
Hägglund, T. (1995). A control-loop performance monitor. Control Engineering Practice,
3, 1543–1551.
11