IEEE TRANSACTIONS ON AUTOMATION SCIENCE AND ENGINEERING, VOL. 11, NO.

4, OCTOBER 2014 1037

Adaptive Observer Based Data-Driven Control for

Nonlinear Discrete-Time Processes
Dezhi Xu, Bin Jiang, Senior Member, IEEE, and Peng Shi, Senior Member, IEEE

Abstract—In this paper, two adaptive observer-based strategies However, these techniques cannot give satisfactory results
are proposed for control of nonlinear processes using input/output when suffering poorly modeling [1]–[3]. This is often the case
(I/O) data. In the two strategies, pseudo-partial derivative (PPD) when dealing with complex, highly nonlinear natural dynamic
parameter of compact form dynamic linearization and PPD vector
of partial form dynamic linearization are all estimated by the adap- processes. Although, adaptive technique based fuzzy logic and
tive observer, which are used to dynamically linearize a nonlinear neural network have been intensively researched. For nonlinear
system. The two proposed control algorithms are only based on the systems in the last two decades [4]–[8]. There are still no assur-
PPD parameter estimation derived online from the I/O data of the ance of high convergence speed, the overheating phenomenon,
controlled system, and Lyapunov-based stability analysis is used avoidance of local minima and so on; meanwhile, there are not
to prove all signals of close-loop control system are bounded. A nu-
merical example, a steam-water heat exchanger example and an general methods to choose the number of the fuzzy rule base
experimental test show that the proposed control algorithm has a and hidden units of common neural network.
very reliable tracking ability and a satisfactory robustness to dis- The term data-driven was first proposed in computer science
turbances and process dynamics variations. and has only recently entered the vocabulary of the control so-
ciety. Because only the input/output (I/O) measurement data is
Note to Practitioners—In an actual industrial process, the dy-
namic behaviors is complex and nonlinear, and their mathematical used in data-driven controller design procedure. The modeling
models are often difficult to obtain. How to design the controller for process, the unmodeled dynamics and the theoretical assump-
unknown nonlinear systems using input/output (I/O) data has be- tions all disappear. So it has caught considerable attention in re-
come one main focus of control researches. Therefore, in this paper, cent years [9]–[17]. There are a few data-driven control (DDC)
two adaptive observer-based data-driven control algorithms are methods as follows: model-free adaptive control (MFAC) [9],
proposed for a class of unknown nonlinear systems. Finally, the ef-
fectiveness of two control strategies are illustrated via simulation [10], virtual reference feedback tuning (VRFT) [11], [12], iter-
study and experimental test. ative learning control (ILC) [13], lazy learning control (LLC)
[14], unfalsified control (UC) methodology [15], dynamic pro-
Index Terms—Adaptive observer, Data-driven control, Lya-
punov-based stability analysis, nonlinear discrete-time systems, gramming method [16], [17] and others [18], [19].
pseudo-partial derivative. As one of the data drive control methods, MFAC has been
proposed and applied in several areas. Hou [9], [10], [20]
has designed MFAC algorithm based on compact form dy-
I. INTRODUCTION
namic linearization (CFDL), partial form dynamic linearization
(PFDL), and full form dynamic linearization (FFDL) for

M ODEL-BASED control techniques are usually imple-

mented under the assumption of superior identification
of process dynamics and their operational circumstances.
single-input single-output (SISO), multi-input single-output
(MISO), and multi-input multi-output (MIMO) systems. How-
ever, the MFAC is still developing. How to prove the stability
and convergence of the tracking problems is one of the open
Manuscript received August 15, 2013; accepted September 04, 2013. Date
of publication November 26, 2013; date of current version October 02, 2014. problems in MFAC [20]. We all know, Lyapunov functional is
This paper was recommended for publication by Associate Editor T. Chen and widely used to analysis the stability of close-loop system [21].
Editor M. C. Zhou upon evaluation of the reviewers’ comments. This work was In this paper, we focus on how to design data-driven con-
supported in part by the National Key Basic Research Program (2011CB710706
and 2012CB215202), the 111 Project (B12018), the National Natural Science troller based on Lyapunov method. Inspired by the work of
Foundation of China (61034005, 61174058, 61273171), Natural Science Foun- dynamic linearization technique of Hou [5], [6], we present
dation of Jiangsu Province (BK20131364), the National Aerospace Science two adaptive observer-based control strategies for nonlinear
Foundation of China (2012ZA52017 and 2011ZA52009), PAPD, the Funding
of Jiangsu Innovation Program for Graduate Education (CXZZ11-0213), processes systems in which the pseudo-partial derivative (PPD)
funding for Outstanding Doctoral Dissertation in NUAA (BCXJ12-04), the theory is used to dynamically linearize the nonlinear system. In
Fundamental Research Funds for the Central Universities, and the Engineering order to achieve the time-varying PPD parameter estimation,
and Physics Sciences Research Council, U.K. (EP/F029195).
D. Xu and B. Jiang are with the College of Automation Engineering, Uni- based on discrete-time adaptive observer technique, a novel
versity of Aeronautics and Astronautics, 210016 Nanjing, China (e-mail: bin- adaptive strategy for computing the PPD term is designed by
[email protected]). using the Lyapunov method. A stability analysis is carried out
P. Shi was with the Department of Computing and Mathematical Sciences,
University of South Wales, CF37 1DL Pontypridd, U.K.. He is now with the to prove that, in the case of perfect parametrization of the CFDL
School of Electrical and Electronic Engineering, University of Adelaide, Ade- and PFDL, the system is globally exponentially stable. Then,
laide, SA 5005, Australia, and the College of Engineering and Science, Victoria inverse control algorithm is used to design the data-driven
University, Melbourne, VIC 8001, Australia (e-mail: [email protected].
au). controller via CFDL, and one-step-ahead weighted predictive
Digital Object Identifier 10.1109/TASE.2013.2284062 control is used to design the controller via PFDL. The stability

1545-5955 © 2013 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

Authorized licensed use limited to: COMSATS INSTITUTE OF INFORMATION TECHNOLOGY. Downloaded on June 02,2022 at 12:25:02 UTC from IEEE Xplore. Restrictions apply.
 1038 IEEE TRANSACTIONS ON AUTOMATION SCIENCE AND ENGINEERING, VOL. 11, NO. 4, OCTOBER 2014

analysis for tracking errors of the proposed algorithms is pro- Assumption 4: The system (1) is generalized Lipschitz,
vided. Last, the paper discusses the different parameter tuning, that is, satisfying , for and
and uses simulations for a numerical plant, a steam-water heat , where , and
exchanger process and experiments to show that the proposed
control algorithm has a good tracking ability and a satisfactory and is a constant.
robustness to disturbances and process dynamics variations. Theorem 2: For the nonlinear system (1), we assume that
The rest of this paper is organized as follows. In Section II, Assumptions 3 and 4 hold. There must exist a parameter vector
a brief descriptions of the CFDL and PFDL is given. In , called PPD vector, system (1) can be transformed into the
Section III, main results of adaptive observer based data-driven following PFDL description when :
control via CFDL are proposed. Next, main results of adaptive
observer based data-driven control via PFDL are developed (3)
in Section IV. Simulation results are presented to show the
effectiveness of the proposed technique in Section V. Finally, where , and
some conclusions are made at end of this paper. Proof: See [5].
The following works are based on basic assumption.
Assumption 5: For Theorems 1 and 2, the norm of the pa-
II. DATA-DRIVEN MODELING (DYNAMICAL rameter and vector is uniformly bounded by
LINEARIZATION TECHNIQUE) constants and , i.e., , and
Consider discrete-time SISO nonlinear system represented in .
the following Nonlinear AutoRegressive with eXogenous input Notice that Assumption 5 can be satisfied for a wide class of
(NARX) model: functions and , by assuming that the output
and the input of the system (1) remains bounded.
(1)
III. CONTROLLER DESIGN VIA COMPACT FORM
where are the system output and input, are the un- DYNAMIC LINEARIZATION
known orders, and is an unknown nonlinear function.
The proposed novel data-driven control algorithm is designed In this section, we will propose a novel data-driven control
with PPD technique. Here, we simply introduce the CFDL and algorithm using the CFDL model. Main contributions in the fol-
PFDL. Details can be seen in [5] and[7]. lowing works include: 1) A novel unknown PPD estimation al-
gorithm; 2) Proposed a data-driven inverse control algorithm;
3) Lyapunov-based stability analysis.
A. Compact Form Dynamic Linearization
The CFDL of the system (1) is based on following two nec- A. Observer-Based PPD Parameter Identification
essary assumptions.
The proposed PPD parameter identification observer has the
Assumption 1: The partial derivative of with respect to
following structure:
the control input is continuous.
Assumption 2: The system (1) is generalized Lipschitz, that
is, satisfying , for and , (4)
where , and
. and is a constant. where is the output estimation error,
Theorem 1: For the nonlinear system (1), we assume that represents an estimate of the PPD parameter, and the gain is
Assumptions 1 and 2 hold. There must exist a parameter , chosen such that in the unit circle.
called PPD, system (1) can be transformed into the following Hence, in view of (2) and (4), the output estimation error dy-
CFDL description when : namics is given by

(2) (5)

where where represents the PPD parameter esti-

Proof: See [5]. mation error.
The PPD term can be tactfully calculated through the estima-
tion of . Namely, an adaptive update law for the parameters
B. Partial Form Dynamic Linearization can be chosen as
The PFDL of the system (1) is based on following two nec-
essary assumptions. (6)
Assumption 3: The partial derivative of with respect to
the control input are continuous. The gain is chosen as follows:
where is a positive constant called control input length con-
stant of linearization for the discrete-time nonlinear system (1).

Authorized licensed use limited to: COMSATS INSTITUTE OF INFORMATION TECHNOLOGY. Downloaded on June 02,2022 at 12:25:02 UTC from IEEE Xplore. Restrictions apply.
 XU et al.: ADAPTIVE OBSERVER BASED DATA-DRIVEN CONTROL FOR NONLINEAR DISCRETE-TIME PROCESSES 1039

where is a positive constant, hence, is positive definite where is reference trajectory. and as given finite pos-
for all . Notice that, by Assumption 5, can be lower itive numbers. Notice that, in many practical systems, because
bounded as their actuators cannot change too fast, the number can be
jammy obtained.
Define observer tracking error , thus

By taking into account (5) and (6), the estimation error dynamics
can be written as
(9)

(7) The robustness of the stability and the performance for data-
driven control law (8) are given in Theorem 4.
where is given by Theorem 4: For given , using the
data-driven control law (8), the solution of close-loop observer
error system (9) is uniformly ultimately bounded (UUB) [22]
The main property of estimate scheme is recapitulated in the for all with ultimate bound ,
following theorem. where is a given positive constant,
Theorem 3: Under Assumption 5, the equilibrium
of the system (7) is globally uniformly stable. Further-
more, the estimation error onverges asymptotically to 0.
Proof: Consider the Lyapunov function

Proof: Define a variable , where for

where are positive constants and is the solution by all . The control law (8) is equivalently expressed as
, the solution exists and is positive definite.
By taking into (7), we have (10)

where

Using (10), (9) becomes

where
. Hence, provided that and
satisfy the following inequalities:

Notice that is negative definite in the variables

. Since in a decreasing and non-negative function,
it converges to a constant value , as , hence,
. This implies that both and remain
bounded for all , and . (11)

B. Controller Design and Stability Analysis Choosing a Lyapunov function as , from (11),
one has
Based on the observer (4), the data-driven inverse control law
can be described as

Since and is bounded, according to the lemma

in [22], using the control law (8), the results of close-loop
observer system (9) are UUB for all with ultimate bound
(8) .

Authorized licensed use limited to: COMSATS INSTITUTE OF INFORMATION TECHNOLOGY. Downloaded on June 02,2022 at 12:25:02 UTC from IEEE Xplore. Restrictions apply.
 1040 IEEE TRANSACTIONS ON AUTOMATION SCIENCE AND ENGINEERING, VOL. 11, NO. 4, OCTOBER 2014

Corollary 1: Under the controller (8), together with the ob- Hence, in view of (3) and (15), the output estimation error dy-
server (4), adaptive laws (6), we can guarantee that the system namics is given by
(1) tracking error is UUB with ultimate
bound . (17)
Proof: Since
where represents the PPD parameter es-
(12) timation error of PFDL.
By taking into account (16) and (17), the estimation error
Taking the absolute value and limiting on both sides of (12), we dynamics can be written as
obtain

(18)
(13)
where is given by
So, the tracking error is UUB for all with ultimate bound
.
Remarks: 1) In order to make the parameter estimation law and denotes the identity matrix.
have a stronger capability in tracking time-varying parameter, a Theorem 5: Under Assumption 5, the equilibrium
retune mechanism which in [7] should be considered as follows: of the system (18) is globally
uniformly stable. Furthermore, the estimation error
converges asymptotically to 0.
(14) Proof: Consider the Lyapunov function
where is a small positive constant and is the initial value
of .
2) If the reference trajectory , we can ob- where are positive constant and is the solution by
tain the , and from Theorem 3, the . . By taking into (18), we have
Hence, the number , we can simply obtained the result
.

IV. CONTROLLER DESIGN VIA PARTIAL FORM

DYNAMIC LINEARIZATION
In this case, the focus is on PFDL, based on the PPD param-
eter vector estimation of adaptive observer which is similar as
Section III-A, we propose a enhanced data-driven control algo- where
rithm. . Hence, provided that
and satisfy the following inequalities:
A. Observer Design
The proposed PPD parameter identification observer has the
following structure:
Notice that is negative definite in the variables
(15) . Since in a decreasing and non-negative function,
it converges to a constant value , as , hence,
where is the output estimation error, . This implies that both and remain
represents an estimate of the PPD parameter vector, and the gain bounded for all , and .
is chosen such that in the unit circle. And the Remark 3: In order to make the parameter estimation law
adaptive update law for the parameters estimate can be have a stronger capability in tracking time-varying parameters, a
chosen as retune mechanism which in [5] should be considered as follows:
(16)
(19)
The gain is chosen as follows:
where is a small positive constant and is the initial value
of .

where is a positive constant, hence, is positive definite B. Enhanced Controller Design

for all . Notice that, by virtue of Assumption 5, can be
The control law can be achieved by minimizing the one-step-
lower bounded as
ahead weighted predictive control performance index

(20)

Authorized licensed use limited to: COMSATS INSTITUTE OF INFORMATION TECHNOLOGY. Downloaded on June 02,2022 at 12:25:02 UTC from IEEE Xplore. Restrictions apply.
 XU et al.: ADAPTIVE OBSERVER BASED DATA-DRIVEN CONTROL FOR NONLINEAR DISCRETE-TIME PROCESSES 1041

denotes the control effort weighting factor. Taking If the parameters satisfy , the control
(15) into (20) and then minimizing it, we can achieve the fol- algorithm given in (24) will be convergent. Suppose the Lya-
lowing control law: punov function is defined as ,
the sufficient condition for the stability of the one-step ahead
data-driven control system is .

V. SIMULATION AND EXPERIMENTAL RESULTS

(21)
In this section, simulation and experiments are shown to
An adaptive modification item (AMI) [23] is introduced demonstrate the validity of the proposed data-driven control
to the control law (21). Then, the control law can be rewritten algorithms. A nonlinear system and steam-water heat ex-
as changer process simulations are conducted to test the algorithm
effectiveness. The output tracking capability, load rejection
capability, and the robustness to process dynamic changes are
tested in this simulation. To demonstrate the real application,
the proposed algorithm is also experimentally tested with a
(22) level control system.

A. Case Study 1
Theorem 6: There exists suitable such that the observer
tracking error will be to 0. The following SISO nonlinear process model is used in the
Proof: Define observer tracking error simulation [26]:
, thus
(25)

Existing the disturbance in input channel which is described as

(23)
The tracking trajectory is given as
where

(26)

So, there exists suitable AMI as For the propose of comparison, the DDC-CFDL and
DDC-PFDL are compared to the neural-based inverse control
(NIC) [27] approach with the disturbance of input.
1) Data-Driven Control Based on CFDL: For the proposed
control law, we choose the sampling time . The pa-
which can make . rameters of proposed control law in Section III are
The AMI, which is time-varying parameter, and may be tuned , and .
for high tracking ability [23], [24]. In most cases, is chosen System responses are shown in Fig. 1(A), which are included
as 1 [24]. [23] gives a simple recursive efficient method, and in output signals of DDC-CFDL and NIC [i.e., Fig. 1A(1)],
this paper we using the recursive efficient method of [23] for input signals [i.e., Fig. 1A(2)], PPD parameter estimation [i.e.,
calculating the in our simulations. Fig. 1A(3)] for the DDC-CFDL in case study 1. The simu-
Corollary 2: Under the controller (22), together with the lations in Fig. 1(A) show good tracking performance, and is
observer (15), adaptive laws (16), we can guarantee that the not affected much by the increasing magnitude of the tracking
tracking error is convergent. trajectory (26). From Fig. 1(A), the tracking error significantly
Proof: The Prove is similar as Corollary 1. decreases using the proposed DDC-CFDL in comparison to
Remark 4: We can also obtain the one-step-ahead control NIC. The proposed data driven controller can achieve a better
from the optimization of the cost function (20) by using the performance in the presence of disturbance.
gradient descent optimizing technique [25], i.e., 2) Data-Driven Control Based on PFDL: The parameters
of proposed control law in Section IV are
(24)
, and .
Fig. 1B(1–3) shows that output signals of DDC-PFDL and
where is the optimizing step. Considering the observer NIC, input signals, and PPD parameter estimation of DDC-
model (15). The sensitivity can be derived PFDL. It can be seen that the proposed DDC-PFDL can achieve
from the a better performance in the presence of disturbance. This is be-
cause the DDC structures does not include a plant model, the
controller is only from I/O data.

Authorized licensed use limited to: COMSATS INSTITUTE OF INFORMATION TECHNOLOGY. Downloaded on June 02,2022 at 12:25:02 UTC from IEEE Xplore. Restrictions apply.
 1042 IEEE TRANSACTIONS ON AUTOMATION SCIENCE AND ENGINEERING, VOL. 11, NO. 4, OCTOBER 2014

Fig. 1. (A) The signals of response for DDC-CFDL in case study 1. (B) The signals of response for DDC-PFDL in case study 1.

B. Case Study 2
Heat exchangers are universal elements in the chemical and
process industry. Temperature control is still a major task if the
heat exchanger is operated over a broad scale. The nonlinear be-
havior depends strongly on the flow rates and on the tempera-
tures of the media. In this section, a steam-water heat exchanger
is considered.
Equations of the steam-water heat exchanger dynamic, which
represent a complex behavior. The experimental equipment is
shown in Fig. 2. The steam condenses in the two-pass shell and
tube heat exchanger, so raising the process water temperature.
The steam flow rate and process water flow rate can be con-
trolled by pneumatic control values. Reference [28] describes
the plant where the heat exchanger dynamic is expressed by
Fig. 2. Heat exchanger.
a Hammerstein model (the nonlinear element follows linear
block) and it is given by:
The control simulations is based on measurement noise 0.005
(27)
rand(1) and the step disturbance 0.1 at 150 s in input channel.
The desired trajectories to be tracked are
(28)

We choose the parameters of DDC-CFDL are

, and . And the The simulation results of two proposed DDC are shown in
parameters of DDC-PFDL are chosen as Fig. 3. Fig. 3A(1) and Fig. 3B(1) give the DDC-CFDL and
and . DDC-PFDL output response. The control signals of two control

Authorized licensed use limited to: COMSATS INSTITUTE OF INFORMATION TECHNOLOGY. Downloaded on June 02,2022 at 12:25:02 UTC from IEEE Xplore. Restrictions apply.
 XU et al.: ADAPTIVE OBSERVER BASED DATA-DRIVEN CONTROL FOR NONLINEAR DISCRETE-TIME PROCESSES 1043

Fig. 3. (A) The signals of response for DDC-CFDL in case study 2. (B) The signals of response for DDC-PFDL in case study 2.

Fig. 4. PCS experimental device.

Fig. 5. Block diagram of level control unit.

laws are shown in Fig. 3A(2) and B(2). Fig. 3A(3) and B(3)
describe the PPD parameter estimations of by two data-driven
controllers. These two simulations show that acceptable
C. Experimental Results
set-point tracking can be obtained with the proposed control
algorithm for the nonlinear process. The results also show that The proposed DDC-CFDL and DDC-PFDL methods were
the close-loop systems with two proposed DDC have stronger experimentally evaluated on a process control system (PCS).
anti-interference ability than robust PID control method [29]. The PCS is made by FESTO company, which is built up by four

Authorized licensed use limited to: COMSATS INSTITUTE OF INFORMATION TECHNOLOGY. Downloaded on June 02,2022 at 12:25:02 UTC from IEEE Xplore. Restrictions apply.
 1044 IEEE TRANSACTIONS ON AUTOMATION SCIENCE AND ENGINEERING, VOL. 11, NO. 4, OCTOBER 2014

VI. CONCLUSION
The PPD was used to dynamically linearize a nonlinear
process, and aggregation was used to predict the PPD. To rely
on dynamically linearize technology, we propose adaptive
observer based two novel and effective data-driven control law
for unknown nonlinear dynamic processes. The two control
has the real-time implementation advantage of not requiring
any iterative computation for determining the control input,
and the Lyapunov-based stability analysis is introduces to the
proposed control systems. The simulations and experiments
show that the proposed control algorithms have good tracking
capability, acceptable robustness to disturbances and process
dynamics changes. It is believed that the proposed algorithm
is a favorable control strategy for unknown nonlinear systems
where real-time application is important.
Furthermore, the DDC for multi-input single-output (MISO),
Fig. 6. Level response to the set-point changes. and multi-input multi-output (MIMO) systems will be studied
in our further works.

REFERENCES
[1] L. Wu, P. Shi, and H. Gao, “State estimation and sliding-mode control
of Markovian jump singular systems,” IEEE Trans. Autom. Control,
vol. 55, no. 5, pp. 1213–1219, May 2010.
[2] L. Wu, C. Wang, and Q. Zeng, “Observer-based sliding mode control
for a class of uncertain nonlinear neutral delay systems,” J. Franklin
Institute, vol. 345, no. 3, pp. 233–253, 2008.
[3] X. Su, L. Wu, P. Shi, and Y. Song, “ model reduction of Takagi-
Sugeno fuzzy stochastic systems,” IEEE Trans. Syst., Man, Cybern.,
Part B: Cybern., vol. 42, no. 6, pp. 1574–1585, Dec. 2012.
[4] S. Tong, P. Shi, and H. Al-Madfai, “Robust fuzzy decentralized con-
trol for nonlinear large scale systems with parametric uncertainties,” J.
Intell. Fuzzy Syst., vol. 19, no. 2, pp. 85–101, 2008.
[5] H. Zhang, L. Cai, and Z. Bien, “A fuzzy basis function vector-based
multivariable adaptive controller for nonlinear systems,” IEEE Trans.
Syst., Man, Cybern., Part B: Cybern., vol. 30, no. 1, pp. 210–217, Feb.
2000.
[6] Q. Zhou, P. Shi, S. Xu, and H. Li, “Observer-based adaptive neural net-
work control for nonlinear stochastic systems with time-delay,” IEEE
Trans. Neural Netw. Learning Syst., vol. 24, no. 1, pp. 71–80, Jan.
Fig. 7. Level response to disturbance. 2013.
[7] X. Liu and Q. Zhang, “New approaches to controller designs based
on fuzzy observers for T-S fuzzy systems via LMI,” Automatica, vol.
39, no. 9, pp. 1571–1582, 2003.
[8] H. Zhang and Y. Quan, “Modeling, identification, and control of a
working units (Level control, temperature control, pressure con- class of nonlinear systems,” IEEE Trans. Fuzzy Syst., vol. 9, no. 2, pp.
trol, flow control). Each unit can work independently, and flex- 349–354, Apr. 2001.
ible connections with other units through hardware and software [9] Z. Hou and S. Jin, “A novel data-driven control approach for a class of
discrete-time nonlinear systems,” IEEE Trans. Control Syst. Technol.,
integration to constitute a complex system. Fig. 4 shows a pho- vol. 19, no. 6, pp. 1549–1558, Nov. 2011.
tograph of this PCS, and the block diagram figure of level con- [10] Z. Hou and S. Jin, “Data driven model-free adaptive control for a class
trol unit is illustrated in Fig. 5, where the SIMENS® S7-300 of MIMO nonlinear discrete-time systems,” IEEE Trans. Neural Netw.,
vol. 22, no. 12, pp. 2173–2188, Dec. 2011.
(lower-computer) is used to collect data for field devices. The [11] A. Sala and A. Esparza, “Extensions to virtual reference feedback
liquid level of tank (B102) is measured by ultrasonic liquid level tuning: A direct method for the design of feedback controllers,”
sensor (LIC102). The control signal is constrained between 0 Automatica, vol. 41, no. 8, pp. 1473–1476, 2005.
[12] F. Previdi et al., “Data-driven control design for neuroprotheses: A vir-
and 10 V for DC motor (P101). tual reference feedback tuning (VRFT) approach,” IEEE Trans. Con-
The parameters of DDC-CFDL are chosen as trol Syst. Technol., vol. 12, no. 1, pp. 176–182, Jan. 2004.
, and . And the [13] J. Xu, Linear and Nonlinear Iterative Learning Control. Berlin, Ger-
many: Springer-Verlag, 2003.
parameters of DDC-PFDL are chosen as [14] T. Pan and S. Li, “A hierarchical search and updating database strategy
and . The experimental for lazy learning,” Int. J. Innovative Comput. Inform. Control, vol. 4,
result is shown in Fig. 6 for the set-point tracking control, and no. 6, pp. 1383–1392, 2008.
[15] G. Battistelli et al., “Stability of unfalsified adaptive switching control
Fig. 7 gives the control responses to a disturbance produced by in noisy environments,” IEEE Trans. Autom. Control, vol. 55, no. 10,
pouring 3 L of water into the tank “B102” at the 200th sample. pp. 2424–2429, Oct. 2010.
These two experiments show that acceptable set-point tracking [16] H. Zhang et al., “Optimal tracking control for a class of nonlinear dis-
crete-time systems with time delays based on heuristic dynamic pro-
and disturbance rejection can be obtained with the proposed gramming,” IEEE Trans. Neural Netw., vol. 22, no. 12, pp. 1851–1862,
control algorithm for the nonlinear process. Dec. 2011.

Authorized licensed use limited to: COMSATS INSTITUTE OF INFORMATION TECHNOLOGY. Downloaded on June 02,2022 at 12:25:02 UTC from IEEE Xplore. Restrictions apply.
 XU et al.: ADAPTIVE OBSERVER BASED DATA-DRIVEN CONTROL FOR NONLINEAR DISCRETE-TIME PROCESSES 1045

[17] H. Zhang et al., Adaptive Dynamic Programming for Control-Algo- Bin Jiang (SM’05) was born in Jiangxi, China, in
rithms and Stability. London, U.K.: Springer-Verlag, 2013. 1966. He received the Ph.D. degree in automatic con-
[18] J. Rojas and R. Vilanova, “Data-driven based IMC control,” Int. J. In- trol from Northeastern University, Shenyang, China,
novative Comput., Inform. Control, vol. 8, no. 3(A), pp. 1557–1574, in 1995.
2012. He has been a postdoctoral fellow, research fellow
[19] C. Li, J. Yi, and T. Wang, “Encoding prior knowledge into data driven and visiting professor in Singapore, France, U.S., and
design of interval type-2 fuzzy logic systems,” Int. J. Innovative Canada, respectively. Now, he is a Chair Professor of
Comput., Inform. Control, vol. 7, no. 3, pp. 1133–1145, 2011. the Cheung Kong Scholar Program in Ministry of Ed-
[20] Z. Hou and Z. Wang, “From model-based control to data-driven con- ucation and Dean of the College of Automation Engi-
trol: Survey, classification and perspective,” Inform. Sci., vol. 235, pp. neering, Nanjing University of Aeronautics and As-
3–35, Jun. 20, 2012. tronautics, China. His research interests include fault
[21] Q. Liu, R. Wang, and D. Wu, “Stability analysis for sampled-data sys- diagnosis and fault tolerant control and their applications.
tems based on multiple Lyapunov functional method,” Int. J. Innova- Prof. Jiang currently serves as an Associate Editor or Editorial Board
tive Comput., Inform. Control, vol. 8, no. 9, pp. 6345–6355, 2012. Member for a number of journals such as the IEEE TRANSACTIONS ON
[22] J. Spooner et al., Stable Adaptive Control and Estimation for Nonlinear CONTROL SYSTEMS TECHNOLOGY, the International Journal of Control,
Systems. New York, NY, USA: Wiley, 2002. the Automation and Systems; Nonlinear Analysis: Hybrid Systems, the In-
[23] Y. Liu, H. Wang, and P. Li, “Kernel learning adaptive one-step-ahead ternational Journal of Applied Mathematics and Computer Science, Acta
predictive control for nonlinear processes,” Asia-Pac. J. Chem. Eng., Automatica Sinica, and the Journal of Astronautics. He is Chair of the Control
vol. 3, pp. 673–679, 2008. Systems Chapter of the IEEE Nanjing Section, a member of the IFAC Technical
[24] F. Gao, F. Wang, and M. Li, “A simple nonlinear controller with Committee on Fault Detection, Supervision, and Safety of Technical Processes.
diagonal recurrent neural network,” Chem. Eng. Sci., vol. 55, pp.
1283–1288, 2000.
[25] Y. Tan and A. Cauwenberghe, “Nonlinear one-step-ahead control using
neural network: Control strategy and stability design,” Automatica, vol. Peng Shi (M’95–SM’98) received the B.Sc. degree
32, no. 12, pp. 1701–1706, 1996. in mathematics from the Harbin Institute of Tech-
[26] K. Narendra and S. Mukhopadhyay, “Adaptive control using neural nology, Harbin, China, the M.E. degree in systems
networks and approximate models,” IEEE Trans. Neural Netw., vol. 8, engineering from Harbin Engineering University,
no. 3, pp. 475–485, May 1997. Harbin, the Ph.D. degree in electrical engineering
[27] J. Labus et al., “Inverse neural network based control strategy for from the University of Newcastle, Australia, the
absorption chillers,” Renewable Energy, vol. 39, no. 1, pp. 471–482, Ph.D. degree in mathematics from the University
2012. of South Australia, and the D.Sc. degree from the
[28] E. Eskinat and S. Johnson, “Use of Hammerstein models in identifica- University of Glamorgan, U.K.
tion of nonlinear systems,” AIChE Journal, vol. 37, no. 2, pp. 255–268, He was a Lecturer at Heilongjiang University,
1991. China, a Postdoctorate and Lecturer at the University
[29] G. Duan and D. Howe, “Robust magnetic bearing control via eigen- of South Australia, a Senior Scientist at the Defence Science and Technology
structure assignment dynamical compensation,” IEEE Trans. Contr. Organization, Australia, and a Professor at the University of Glamorgan (now
Syst. Technol., vol. 11, no. 2, pp. 204–215, 2003. The University of South Wales), U.K. Now, he is a Professor with the Uni-
versity of Adelaide and Victoria University, Australia. His research interests
include system and control theory, computational intelligence, and operational
research.
Dr. Shi is a Fellow of the Institution of Engineering and Technology (U.K.),
and a Fellow of the Institute of Mathematics and its Applications (U.K.). He has
Dezhi Xu was born in Jiangsu, China, in 1985. He been on the editorial board of a number of journals, including AUTOMATICA,
received the M.Sc. degree in automatic control from the IEEE TRANSACTIONS ON AUTOMATIC CONTROL, the IEEE TRANSACTIONS
Lanzhou University of Technology, Lanzhou, China, ON FUZZY SYSTEMS, the IEEE TRANSACTIONS ON CYBERNETICS, the IEEE
in 2010. Currently, he is working towards the Ph.D. TRANSACTIONS ON CIRCUITS AND SYSTEMS-I, and IEEE ACCESS.
degree at the College of Automation Engineering,
Nanjing University of Aeronautics and Astronautics,
Nanjing, China.
His research interests include fault diagnosis and
fault-tolerant control, data-driven control.

Authorized licensed use limited to: COMSATS INSTITUTE OF INFORMATION TECHNOLOGY. Downloaded on June 02,2022 at 12:25:02 UTC from IEEE Xplore. Restrictions apply.