ISA Transactions
journal homepage: www.elsevier.com/locate/isatrans
Research article
art ic l e i nf o a b s t r a c t
Article history: An improved proportional-integral-derivative (PID) controller based on predictive functional control
Received 19 August 2016 (PFC) is proposed and tested on the chamber pressure in an industrial coke furnace. The proposed design
Received in revised form is motivated by the fact that PID controllers for industrial processes with time delay may not achieve the
31 October 2016
desired control performance because of the unavoidable model/plant mismatches, while model pre-
Accepted 15 November 2016
dictive control (MPC) is suitable for such situations. In this paper, PID control and PFC algorithm are
combined to form a new PID controller that has the basic characteristic of PFC algorithm and at the same
Keywords: time, the simple structure of traditional PID controller. The proposed controller was tested in terms of
PID control set-point tracking and disturbance rejection, where the obtained results showed that the proposed
Predictive functional control
controller had the better ensemble performance compared with traditional PID controllers.
time delay processes
& 2016 ISA. Published by Elsevier Ltd. All rights reserved.
ensemble performance
1. Introduction parameters tuning method for rst-order plus dead time pro-
cesses (FOPDT) [11]. Later, internal model control based PID
Coke furnace plays an important role in the petroleum re- tuning method was proposed to improve the dynamic re-
ning industry by providing different oil products for in- sponses of the closed-loop system [12]. Nowadays, substantial
dustrial needs [14]. Since the yield, quality and safety of the research results on PID tuning for different process models are
process are associated with proper operation, modeling based available [1317].
control of relevant process variables are very important [5,6]. However, for industrial coke processes with long time delay
However, due to the interactions in the whole coke process and and nonlinearity, the application of the aforementioned traditional
batch operation, accurate modeling may result in complex PID controllers may not achieve the desired effect [1821]. In order
model structure or higher order and poses some complexity for to solve this problem, smith predictor was proposed and known as
the subsequent controller design [7,8]. From control point of an advanced PID control algorithm to deal with the time delay
view, PID controllers are the most common ones with wide- processes [2226]. Unlike PID controllers, smith predictor converts
the closed-loop system transfer function into a formula in which
spread applications. It is well-known that good performance of
time delay is separated alone, which results in the fact that the
a PID controller depends on the tuning of its parameters,
response curve of the entire closed-loop system is the same as a
therefore, a lot of tuning methods have been proposed. Ziegler
corresponding closed-loop system without time delay. However, it
and Nichols rst proposed the famous tuning guidelines called
is not very easy to enhance process performance with smith pre-
Z-N method and had been referred to numerous times in
dictor due to the fact that accurate process models cannot be ob-
practical applications [9]. In order to compensate for the in-
tained for industrial processes.
sufciency of Z-N method in dealing with time delay processes, With the development of computer control technologies,
Cohen and Coon proposed an improved parameters tuning model predictive control (MPC) was proposed as an effective
approach [10]. Tyreus and Luyben came up with a special advanced algorithm and signicantly researched in handling
chemical processes with time delays, together with the relevant
A novel PID controller optimized by predictive functional control (PFC) is pro- issues of input-output modeling and state space model identi-
posed and tested on chamber pressure in the industrial coke furnace in this paper. cation or estimation [2740]. MPC is successful because it
The resulting controller displays the performance of both PFC and PID control with bears the prediction and an effective feedback correction me-
easy implementation in practice. The performance of the proposed PID controller is chanism in designing a control law, which can consider the
compared with traditional PID controller, from which results show that the pro-
posed PID controller provides better performance than traditional PID controller.
complex process behavior with uncertainties. However, How-
E-mail address: jmzhang@csc.zju.edu.cn ever, limited by the cost, hardware and so on, the application of
1
Tel.: +86-571-87952233. MPC is more troublesome than PID, which makes it necessary to
http://dx.doi.org/10.1016/j.isatra.2016.11.006
0019-0578/& 2016 ISA. Published by Elsevier Ltd. All rights reserved.
Please cite this article as: Zhang J. Design of a new PID controller using predictive functional control optimization for chamber pressure
in a coke furnace. ISA Transactions (2016), http://dx.doi.org/10.1016/j.isatra.2016.11.006i
2 J. Zhang / ISA Transactions ()
nd a compromise between MPC and PID. Some representative parameters of the PID controller, where the obtained PID con-
strategies are as follows. Xu, Li, and Cai proposed a novel PID troller has the same performance as PFC algorithm and the simple
controller that was optimized by generalized predictive control structure as traditional PID controllers simultaneously.
(GPC) and obtained the similar performance as GPC [41], how- The aforementioned process model can be transformed into a
ever, the resultant PID is obtained through linear approxima- corresponding differenced equation model by adding a zero-order
tion. A novel multivariable predictive fuzzy-PID control system holder at the sample time Ts .
was developed by incorporating the fuzzy and PID control ap-
proaches into the predictive control framework by Savran [42]. ym (k ) = am ym (k 1) + K (1 am ) u (k 1 L ) (2)
Lee and Yeo developed a new PID controller on the basis of
where ym (k ) is the output of the process model at time instant k ,
simplied GPC [43]. Many other advanced control algorithms
am = eTs/ T , u (k 1 L ) is the input of the process model at time
have also been introduced to improve the performance of tra-
instant k 1 L , L is the time delay with L = /Ts .
ditional PID controllers in dealing with time delay processes
Based on Eq. (2) and to obtain the performance of PFC, we need
[4448].
to construct a process model without time delay, which results in
In this study, a predictive functional control (PFC) algorithm is
used to derive a new PID controller, which has the excellent the following formulation
performance of PFC algorithm and simultaneously, the same ymav (k ) = am ymav (k 1) + K (1 am ) u (k 1) (3)
structure as traditional PID controllers. The chamber pressure in
a coke furnace is considered as a case study, where the well- where ymav (k ) is the output of the process model that is free of
known PID controllers tuned by the Z-N method, T-L method and time delay at time instant k .
IMC method are used for comparison with the proposed PID The output prediction based on Eq. (3) after P steps is
controller.
ymav (k + P ) = am P ymav (k ) + K (1 am P ) u (k ) (4)
Please cite this article as: Zhang J. Design of a new PID controller using predictive functional control optimization for chamber pressure
in a coke furnace. ISA Transactions (2016), http://dx.doi.org/10.1016/j.isatra.2016.11.006i
J. Zhang / ISA Transactions () 3
FRC8107 FRC8108
43.5 t/h 43.6 t/h
Table 2
Circulating oil from pumps
102/1,2,3
Tuning parameters for servo problem.
then Kp (k ) , Ki (k ) , Kd (k ) are obtained and the nal PID controller is implemented on the process. The
Kp (k ) = w2 (k ) 2Kd (k ) advantages are that the process controller is designed based ad-
vanced control using PFC theory and three PID parameters are
Ki (k ) = w1 (k ) KP (k ) Kd (k )
obtained but not xed through optimization for the controller to
Kd (k ) = w3 (k ) (10) achieve improved control performance.
It can be easily seen that w will be innite if (k ) E (k ) is ap- E
proximating zero, that is, Kp (k ) , Ki (k ) , Kd (k ) will be innite and is
unrealistic for the PID controller when the control system has 4. Case study: furnace pressure system control in the coke
reached the steady state. Thus it is necessary to set a small per- furnace
missible error limitation , where Kp (k ) , Ki (k ) , Kd (k ) remain un-
changed as the previous sampling instant 4.1. Coke unit description
Kd (k ) = w3 (k ) (10a) The two branches add together and go to the coke towers (T101/
5,6) for coke removing nally.
The control input at sampling instant k is
+ Kd (k )(e1 (k ) 2e1 (k 1) + e1 (k 2)) (11) The chamber pressure in the coke furnace is one of the im-
At each time instant, if e1 (k ) , Kp (k ) , Ki (k ) , Kd (k ) will keep the portant output variables. Here the manipulated variable is the
same value as the previous time instant. If e1 (k ) > , w (k ) is obtained opening the ue damper and the steady state operating point is
by Eq. (9), then Kp (k ) , Ki (k ) , Kd (k ) can be acquired form Eq. (10). summarized in Table 1.
Remark 1. It is shown that the proposed PID parameters are op- 4.3. The process model
timized, where in fact a PFC is designed for the considered process
Under the conditions of Table 1, the process model is roughly
modeled as
Please cite this article as: Zhang J. Design of a new PID controller using predictive functional control optimization for chamber pressure
in a coke furnace. ISA Transactions (2016), http://dx.doi.org/10.1016/j.isatra.2016.11.006i
4 J. Zhang / ISA Transactions ()
y(k)
1 1
0.5 0.5
0 0
0 50 100 150 200 0 50 100 150 200
k k
200 200
PFC-PID PFC-PID
IMC-PID IMC-PID
100 Z-N 100 Z-N
T-L T-L
0 0
u(k)
u(k)
-100 -100
-200 -200
-300 -300
-400 -400
0 50 100 150 200 0 50 100 150 200
k k
Fig. 2. (a) Closed-loop servo response under output disturbance for Case 1 (sampling time Ts 20 s). (b) Closed-loop servo response under input disturbance for Case 1
(sampling time Ts 20 s).
1 1
0.5 0.5
0 0
0 50 100 150 200 0 50 100 150 200
k k
200 200
PFC-PID PFC-PID
IMC-PID IMC-PID
100 Z-N 100 Z-N
T-L T-L
0 0
u(k)
u(k)
-100 -100
-200 -200
-300 -300
-400 -400
0 50 100 150 200 0 50 100 150 200
k k
Fig. 3. (a) Closed-loop servo response under output disturbance for Case 2 (sampling time Ts 20 s). (b) Closed-loop servo response under input disturbance for Case 2
(sampling time Ts 20 s).
Please cite this article as: Zhang J. Design of a new PID controller using predictive functional control optimization for chamber pressure
in a coke furnace. ISA Transactions (2016), http://dx.doi.org/10.1016/j.isatra.2016.11.006i
J. Zhang / ISA Transactions () 5
y(k) 1 1
y(k)
0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2
0 0
0 50 100 150 200 0 50 100 150 200
k k
300 300
PFC-PID
IMC-PID PFC-PID
200 200
Z-N IMC-PID
T-L Z-N
100 100 T-L
0 0
u(k)
u(k)
-100 -100
-200 -200
-300 -300
-400 -400
0 50 100 150 200 0 50 100 150 200
k k
Fig. 4. (a) Closed-loop servo response under output disturbance for Case 3 (sampling time Ts 20 s). (b) Closed-loop servo response under input disturbance for Case 3
(sampling time Ts 20 s).
Table 3 Table 4
Statistical results for each method. Mean tracking error (MTE) for each method.
Case item Overshoot (%) Response time (s) Settling time (s) Case item Output disturbance Input disturbance
Please cite this article as: Zhang J. Design of a new PID controller using predictive functional control optimization for chamber pressure
in a coke furnace. ISA Transactions (2016), http://dx.doi.org/10.1016/j.isatra.2016.11.006i
6 J. Zhang / ISA Transactions ()
to the output at time instant k = 100 respectively. Here if the [10] Cohen GH, Coon GA. Theoretical investigation of retarded control. Trans ASME
tracking error is small enough and acceptable, we think that the 1953;75:82734.
[11] Tyreus BD, Luyben WL. Tuning PI controllers for integrator/dead time pro-
controlled system is indeed well controlled. Compared with the cesses. Trans ASME 1992;31(11):262831.
set-point (here selected as 1), we deem that the permissible error [12] Bequette BW. Process control: modeling, design and simulation.Upper Saddle
as 10 4 is small enough. It means that the parameters of the River, NJ: Prentice Hall; 2003.
[13] Rivera D, Morari M, Skogestad S. Internal model control 4. PID controller de-
proposed PID controller is invariable and remain the same as the sign. Ind. Eng. Chem. Process Des Dev 1986;25:25265.
previous sampling instant when the absolute value of the error [14] Venkatashankar V, Chidambaram M. Design of P and PI controllers for un-
between actual chamber pressure value and set-point is less-than stable FOPTD model. Int J Control 1994;60(1):13744.
[15] Luyben WL. Tunning proportional-integral-derivative controllers for in-
10 4. The parameters of PFC algorithm and the traditional PID
tegrator/dead time processes. Ind Eng Chem Res 1996;35(10):34803.
controllers are shown in Table 2. [16] Lee Y, Lee M, Park S, Brosilow C. PID controller tuning for desired closed-loop
Figs. 24 show the responses of three cases. The ensemble responses for SISO systems. AIChE J 1998;44(1):10616.
[17] Padma R, Srinivas MN, Chidambaram M. A simple method of tuning PID controllers
performance of the proposed PID controller is better than the
for stable and unstable FOPTD systems. Comput Chem Eng 2004;28(11):220118.
other three PID controllers. In Fig. 2, we can see that the responses [18] Zhang R, Cao Z, Li P, Gao F. Design and implementation of an improved linear
of proposed method and IMC method are smoother than the other quadratic regulation control for oxygen content in a coke furnace. IET Control
two methods that show more drastic oscillations. Specically, the Theory Appl 2014;8(14):130311.
[19] Zhang RD, Wang SQ. Support vector machine based predictive functional
response of Z-N method shows an unacceptable overshoot, control design for output temperature of coking furnace. J Process Control
whereas the tracking performance of T-L method is worse than the 2008;18(5):43948.
other three methods. Compared with the proposed method, the [20] Garpinger O, Hgglund T, strm KJ. Performance and robustness trade-offs in
PID control. J Process Control 2014;24(5):56877.
performance of IMC method is inferior because its curve shows a [21] Miccio M, Cosenza B. Control of a distillation column by type-2 and type-1
bigger overshoot and a more drastic oscillation. In Figs. 34, the fuzzy logic PID controllers. J Process Control 2014;24(5):47584.
situation is the same as that in Fig. 2, where the performance of [22] Palmor Z. Stability properties of Smith dead-time compensator controllers. Int
J Control 1980;32(6):93749.
proposed method is better than the other three methods. Here Z-N [23] Boudjehem D, Sedraoui M, Boudjehem B. A fractional model for robust frac-
method shows an unsatisfactory overshoot and oscillation which tional order Smith predictor. Nonlinear Dyn 2013;73(3):155763.
is more obvious in Fig.4, T-L method presents the bad tracking [24] Wu S, Zhang R, Lu R, Gao F. Design of dynamic matrix control based PID for
residual oil outlet temperature in a coke furnace. Chemom Intell Lab Syst
performance, whereas performance of the IMC method still loses
2014;134:1107.
to the proposed method in control performance. [25] Tsai MH, Tung PC. Modied Smith predictor with a robust disturbance re-
Table 3 shows the statistical results of overshoot, response time duction scheme for linear systems with small time delays. EXPERT SYSTEMS
and settling time for each method. We can easily nd that the 2012;29(4):394410.
[26] Vanavil B, Uma S, Rao AS. Smith predictor based parallel cascade control
ensemble control performance of the proposed PFC-PID control strategy for unstable processes with application to a continuous bioreactor.
strategy is the best. The settling time of the proposed method is Chem Product Process Model 2012;7(1):22.
the smallest under all cases. [27] Qin SJ, Badgwell TA. A survey of industrial model predictive control technol-
ogy. Control Eng Pract 2003;11(7):73364.
The statistical results of mean tracking error (MTE) for each method [28] Cao Z, Yang Y, Lu J, Gao F. Two-time-dimensional model predictive control of
are shown in Table 4. It is obvious that the MTE of the proposed PID weld line positioning in bi-Injection molding. Ind Eng Chem Res 2015;54
controller is the smallest, which veries the aforementioned results. (17):4795804.
[29] Zhang R, Gao F. An improved decoupling structure based state space MPC
design with improved performance. Syst Control Lett 2015;75:7783.
[30] le Roux JD, Padhi R, Craig IK. Optimal control of grinding mill circuit using
5. Conclusion model predictive static programming: a new nonlinear MPC paradigm. J
Process Control 2014;24(12):2940.
[31] Cao Z, Yang Y, Lu J, Gao F. Constrained two dimensional recursive least squares
An improved PID controller optimized by PFC algorithm in this model identication for batch processes. J Process Control 2014;24(6):8719.
paper. The case study on the chamber pressure of a coke furnace [32] Zou Q, Jin Q, Zhang R. Design of fractional order predictive functional control
for fractional industrial processes. Chemom Intell Lab Syst 2016;152:3441.
shows that the proposed PID controller shows improved perfor-
[33] Cole WJ, Morton DP, Edgar TF. Optimal electricity rate structures for peak
mance than traditional PID controllers under conditions of various demand reduction using economic model predictive control. J Process Control
disturbances and model/plant mismatches, where simultaneously 2014;24(8):13117.
has the same structure as PID controller. [34] Cao Z, Zhang R, Yang Y, Lu J, Gao F. Discrete-time robust iterative learning
Kalman ltering for repetitive processes. IEEE Trans Autom Control 2016;61
(1):2705.
[35] Teh PS, Trinh H. Design of unknown input functional observers for nonlinear
References systems with application to fault diagnosis. J Process Control 2014;23:116984.
[36] Zhang R, Lu R, Xue A, Gao F. New minmax linear quadratic fault-tolerant
tracking control for batch processes. IEEE Trans Autom Control 2016;61
[1] Sawarkar AN, Pandit AB, Samant SD, Joshi JB. Petroleum residue upgrading via (10):304551.
delayed coking: a review. Can J Chem Eng 2007;85(2):124. [37] Tadic P, Durovic Z. Particle ltering for sensor fault diagnosis and identica-
[2] Haseloff V, Friedman YZ, Goodhart SG. Implementing coker advanced process tion in nonlinear plants. J Process Control 2014;24:4019.
control. Hydrocarb Process 2007;86(6):99103. [38] Cao Z, Lu J, Zhang R, Gao F. Iterative learning Kalman lter for repetitive
[3] Zhang R, Cao Z, Lu R, Li P, Gao F. State-space predictive-p control for liquid processes. J Process Control 2016;46:92104.
level in an industrial coke fractionation tower. IEEE Trans Autom Sci Eng [39] Hong JJ, Zhang J, Morris J. Progressive multi-block modelling for enhanced
2015;12(4):151624. fault isolation in batch processes. J Process Control 2014;24:1326.
[4] Zhang R, Xue A, Gao F. Temperature control of industrial coke furnace using novel [40] Zhang R, Lu J, Qu H, Gao F. State space model predictive fault-tolerant control
state space model predictive control. IEEE Trans Ind Inform 2014;10(4):208492. for batch processes with partial actuator failure. J Process Control 2014;24
[5] Cao Z, Yang Y, Yi H, Gao F. Priori knowledge based online closed-loop iden- (5):61320.
tication for batch process with an application to injection molding. Ind Eng [41] Xu M, Li SY, Cai WJ. Practical receding-horizon optimization control of the air
Chem Res 2016;55(32):881829. handling unit in HVAC systems. Ind Eng Chem Res 2005;44(8):284855.
[6] Zhang WY, Huang DX, Wang YD, Wang JC. Adaptive state feedback predictive [42] Savran A. A multivariable predictive fuzzy PID control system. APPLIED SOFT
control and expert control for a delayed coking furnace. Chin J Chem Eng COMPUTING 2013;13(5):265867.
2008;16(4):5908. [43] Lee KN, Yeo YK. Predictive PID tuning method based on the simplied GPC
[7] Cao Z, Zhang R, Lu J, Gao F. Two time dimensional recursive system identi- control law. J Chem Eng Jpn 2009;42(4):27480.
cation incorporating priori pole and zero knowledge. J Process Control [44] Saad MS, Jamaluddin H, Darus IZM. PID controller tuning using evolutionary
2016;39:10010. algorithms. WSEAS Trans Syst Control 2012;7(4):13949.
[8] Cao Z, Zhang R, Lu J, Gao F. Online identication for batch processes in closed [45] Zhang R, Cao Z, Bo C, Li P, Gao F. New PID controller design using extended
loop incorporating priori controller knowledge. Comput Chem Eng nonminimal state space model based predictive functional control structure.
2016;90:22233. Ind Eng Chem Res 2014;52(8):323645.
[9] Ziegler J, Nichols N. Optimum settings for automatic controllers, 64. Trans. [46] Ribic AI, Matausek MR. A dead-time compensating PID controller structure
ASME; 1942. p. 75968. and robust tuning. J Process Control 2012;22(7):13409.
Please cite this article as: Zhang J. Design of a new PID controller using predictive functional control optimization for chamber pressure
in a coke furnace. ISA Transactions (2016), http://dx.doi.org/10.1016/j.isatra.2016.11.006i
J. Zhang / ISA Transactions () 7
[47] Papadopoulos KG, Margaris NI. Optimal automatic tuning of active damping improved GA optimization for oxygen content in a coke furnace. Ind Eng Chem
PID regulators. J Process Control 2013;23(6):90515. Res 2016;55(22):646574.
[48] Slavov T, Roeva O. Application of genetic algorithm to tuning a PID controller [50] Zhang RD, Xue AK, Lu RQ, Li P, Gao FR. Real-time implementation of improved
for glucose concentration control. WSEAS Trans Syst 2012;11(7):22333. state space MPC for air-supply in a coke furnace. IEEE Trans Ind Electron
[49] Zhang R, Tao J, Gao F. A new approach of Takagi-Sugeno fuzzy modeling using 2014;61(7):35329.
Please cite this article as: Zhang J. Design of a new PID controller using predictive functional control optimization for chamber pressure
in a coke furnace. ISA Transactions (2016), http://dx.doi.org/10.1016/j.isatra.2016.11.006i