You are on page 1of 5

2013 Fifth International Conference on Intelligent Human-Machine Systems and Cybernetics

The Research Survey of System Identification Method

Li Fu Pengfei Li
Automation department Automation department
Shenyang Aerospace University Shenyang Aerospace University
Shenyang, China Shenyang, China
e-mail: ffulli@yahoo.com.cn e-mail: lipengfei1988417@163.com.cn

Abstract—At present, as a method of establishing mathematical identification is in accordance with a criterion and finally we
model of the system, the system identification has been widely select a model which fits the data best from a set of models.
applied to the automatic control, aviation, spaceflight, However, a Chinese scholar called Ding Feng [4] on system
astronomy, medicine, biology, marine ecology and society, identification gives a new definition that system
economics and many other fields. With the rapid development identification consists of designing an appropriate input
of science and technology, the status of system identification signal, using experimental input and output data, selecting a
technique in various disciplines is becoming increasingly class of models, constructing an error criterion function, and
important. This paper firstly introduces traditional methods of finally determining a model that fits the data best by
linear system identification, and then modern methods of
optimization methods. This comprehensive definition
nonlinear system identification are introduced briefly based on
emphasizes the design of the input signal strongly in system
the neural network, fuzzy logic, genetic algorithm, swarm
intelligence optimization algorithms, auxiliary model identification and optimization methods, and the experiment
identification algorithm, multi-innovation identification design of system identification accords with reality all the
algorithm and hierarchical identification algorithm, and finally more.
the author analyses the developing tendency and prospect of This paper firstly introduces traditional methods of linear
system identification. system identification and points out their advantages and
disadvantages, and then modern methods of nonlinear
Keywords-System identification, neural network, fuzzy logic, system identification are introduced briefly based on the
genetic algorithm, swarm intelligence optimization neural network, fuzzy logic, genetic algorithm, swarm
algorithms,auxiliary model identification algorithm, multi- intelligence optimization algorithms, auxiliary model
innovation identification algorithm, hierarchical identification identification algorithm, multi-innovation identification
algorithm algorithm and hierarchical identification algorithm, and
finally the author analyses the developing tendency and
I. INTRODUCTION prospect of system identification.
System identification is a method of using the input and
output data of the system to build the mathematical model of II. TRADITIONAL METHODS FOR LINEAR SYSTEM
the system, which is being widely applied to various fields of
IDENTIFICATION
production and life. Since modern systems are becoming
more and more complex, and even their internal mechanisms System identification includes linear system
are unclear in whole, therefore, whether or not researchers identification and nonlinear system identification. The
design a rational system identification method to establish traditional identification methods [4-6] are on the basis of
the mathematical model of the system plays an important determining the system model framework and the error
role in the process of system identification. criterion function, and then according to input and output
System identification has been developed for dozens of data of the system, the model parameters are determined by
years during which different scholars have different views on minimizing the error criterion function. Traditional
the definition of identification. In 1962, L. A. Zadeh [1] identification methods include least squares method, gradient
thought that identification was on the basis of the input and correction method and maximum likelihood method and so
output data in order that we could determine an equivalent on. The traditional identification methods have the following
model from a group of given models for the system under advantages:
test. And, Subsequently, P. Eykhoff [2] in 1974, further • Generally speaking, we often take the generalized
defined the concept of identification: " Identification error criterion function as the criterion function, and
problem can be summed up as a kind of calculus which we need to take account of the impact of system
represents the essential characteristics of the objective noise in the process of identification.
system by a model, and the understanding of the objective • Input signal is the Gaussian white noise in theory,
system can be expressed as useful forms by using this model. but we often use the pseudo random signal.
ʾThe definition of identification proposed by L. Ljung [3] • These methods are suitable for online identification.
in 1978 is more specific and it is said that identification However, these methods also have the following
contains three elements: data, models and standards and that disadvantages at the same time:

978-0-7695-5011-4/13 $26.00 © 2013 IEEE 395


397
DOI 10.1109/IHMSC.2013.242
• The least square identification method is under the excitation function of the hidden nodes, and take the dilation
condition that the input signal must be known and and translation parameters of the wavelet function as the
that the input signal need to change relatively and weights and threshold value from input layer to hidden layer.
richly, but system cannot satisfy this condition in The wavelet neural network has the following advantages:
some dynamic systems. • The element and structure of wavelet network is
• Researchers can get a good identification result by determined by wavelet theory, so it can avoid the
these methods for the linear system, but not for the blindness of structure design such as the BP neural
nonlinear system. network.
• Structure identification and parameter identification • The wavelet neural network has good learning
cannot be synchronized. capacity and high approximation precision.
• The wavelet neural network has simple structures
III. MODERN METHODS FOR NONLINEAR SYSTEM and high convergence speed and can avoid the local
IDENTIFICATION convergence state of the system.
There have been many modern methods for nonlinear In the realm of system identification, wavelet neural
system identification in recent years based on the neural network has more and more application potential especially
networks, fuzzy logic, genetic algorithm, swarm intelligence in nonlinear system identification, and it proposes a new and
optimization algorithms, auxiliary model identification effective method for uncertainty and complex nonlinear
algorithm, multi-innovation identification algorithm and system identification, and has broad application prospects
hierarchical identification algorithm [4-7, 35] etc. The [13-20].
following is a brief look at these methods: B. System Identification Method Based on Fuzzy Logic
A. System Identification Method Based on Neural Network Theory
Artificial neural network is a high developed technology In recent years, fuzzy logic theory has made great
begun from the end of the 20th century. So far, the progress in solving the problem of nonlinear system
application area of neural network extends to all kinds of identification. Fuzzy identification needs to solve the
areas, especially in intelligent control, nonlinear problems such as space division of input-output,
optimization, computer vision, biomedical engineering, robot determination of the input variables and membership
and system identification where researchers have achieved function, structure determination of output space function
significant achievements. Neural network identification is and extraction of fuzzy rules. Fuzzy identification has many
based on the neural network structure of nonlinear system, advantages: this method can effectively identify large time-
can simulate the input-output relation by using the ability of varying delay system and multi-input single-output system,
neural network which can infinite approach to any nonlinear can identify the complex system with morbid structure, and
mapping, can realize the simple learning algorithm on can obtain the controlled object’s combination model of
engineering by using the ability of self-adaption and self- quantitative and qualitative. There are many fuzzy
learning of the neural network, and finally obtains the identification methods, and the typical identification methods
forward or inverse model of the system through training. are as follows. C. Sun solves the problem of structure
This method has the following characteristics: identification by the method based on K-D tree [21]. Fuzzy
• No need to establish the model structure of actual clustering methods which have different characteristics have
system, because the neural network itself is a model been used widely in structure identification for fuzzy model
of the network identification. [22]. Fuzzy neural network has been widely used in
• Any linear or nonlinear model can be identified. identification and control for nonlinear system [23]. Genetic
• Not only the neural network is a model, but also it is algorithm can be used in fuzzy identification in various
an actual system which can be achieved by physics. ways, from the simplest model parameter identification to
It also has the following limitations at the same time: the most complex learning of rules geometry [24, 25]. In
• There is a local minimum problem. addition, there are also various fuzzy identification methods
[26, 27].
• Long training time and slow learning speed.
• It is difficult to extract ideal training samples C. System Identification Method Based on Genetic
• It is not easy to optimize the network structure. Algorithm
• It is difficult to solve the convergence problem Genetic algorithm is a kind of bionic optimization
completely for the neural network algorithm in algorithm. Genetic algorithm is a self-adaptive, heuristic and
theoretically. global optimization algorithm which is formed by abstracting
Many scholars have proposed some improved neural and simplifying the biological evolution process, and is
network identification methods and have achieved good based on nature selection and heredity. The characteristics of
identification effects [8-12]. system identification method based on genetic algorithm are
In the last decade, applying the wavelet neural network as follows:
for system identification is a new method of neural network • GA is a global search algorithm, and its convergence
identification. The network structure is similar to RBF neural speed is fast.
network, and we let the wavelet function replace the

398
396
• Don’t need priori knowledge of system to identify identification methods and have achieved good identification
the nonlinear system. effects [37-42].
• It isn’t a problem for GA to select the initial value,
E. Auxiliary Model Identification Method
and even though initial value is far away from the
optimal solution, after several generations Auxiliary model identification [43] is a new
optimization, system can obtain optimal solution. identification method with which we can use measurable
Meanwhile, this method has some limitations: genetic information of the system to establish an auxiliary model.
algorithm has local minimum problem and premature Then the output of the model is used to replace unmeasured
convergence problem, and it is mainly used in parameter variables of the system and the output of the auxiliary model
identification. In addition, GA can not synchronous identify is more close to the unmeasured variables through selecting
the structure and parameter of the system. In [28], Li and the auxiliary model parameter. Finally, consistent estimation
Sun propose a system identification method by applying the of the system parameters is achieved. It was during 2004 that
improved genetic algorithm, and finally obtains the unbiased the auxiliary model identification algorithm was presented in
estimator of system parameter. In [29], Li proposes which the field of international control [44]. Nowadays, this method
combines genetic algorithm and evolutionary programming, has been widely applied to solving the identification problem
and this method can identify the structure and parameter of of the dual-rate sampled data systems [44] and the systems
the system at the same time. In [30], Jiang and Wang with irregularly missing data [45]. Auxiliary model
propose a parameter estimation method to nonlinear system identification has become a kind of identification methods
model based on genetic algorithm, and finally obtains a good and presently is used in the study on adaptive signal
identification effect. processing, adaptive identification and parameter estimation
problems of the linear and nonlinear models. These methods
D. System Identification Method Based on Swarm are as follows: recursive least squares identification method
Intelligence Optimization Algorithms for the auxiliary model of the system with irregularly
In recent years, swarm intelligence and bionic calculation missing data [45], recursive least squares identification
[31] has become an important area of high-tech research in method for the auxiliary model of the multi-rate multi-input
contemporary era, and has been widely used in such realms systems [46], auxiliary model-based least-squares
as artificial intelligence, robot, and data analysis. Swarm identification methods for Hammerstein output-error systems
intelligence optimization algorithms are a string of [47] and so on.
optimization algorithms, through simulating all kinds of F. Multi-innovation Identification Method
group behavior of creatures of nature, and using mutual
communication and cooperation between individuals in the Multi-innovation identification is an important branch of
group to achieve search. At present, swarm intelligence the system identification. The innovation is useful
optimization algorithms mainly include ant colony information which is able to improve the parameters
optimization algorithm [32], particle swarm optimization estimation precision or accuracy of state estimation. Multi-
algorithm [33], flora optimization algorithm [34], artificial innovation identification method [48] is an identification
bee colony optimization algorithm [35] and artificial fish method in which we can fully utilize and extend the
swarm optimization algorithm [36]. These algorithms have identification innovation. In other words, this identification
many advantages as follows: the optimization process is a theory is based on the innovation by extending identification
global optimization process, and it does not depend on initial innovation, turning scalar innovation to innovation vector
value of the algorithm; optimization have wide application, and changing innovation vector to innovation matrix. Multi-
and it needn’t priori knowledge of the optimization problem; innovation identification has become a new identification
the algorithm itself is simple and easy to implement, has field and it can be applied to the study on parameter
standardized procedure, and integrates with other algorithm estimation, adaptive signal processing and adaptive
easily; the algorithm is based on swarm intelligence theory identification problems of all kinds of models. These
and has a good potential parallelism. methods are as follows: multi-innovation stochastic gradient
In essence, system identification is an optimization identification method [49], multi-innovation least squares
problem, thus all kinds of swarm intelligence optimization identification methods based on the auxiliary model for
algorithms can be used in system identification, and different MISO systems [50], multi-innovation stochastic gradient
optimization algorithms constitute different identification identification algorithm based on feed-forward neural
algorithms. According to this theme, we can transform the networks [51] and so on.
parameter identification problem into parameter optimization G. Hierarchical Identification Method
problem under certain evaluation function, map the
parameters to be identified to ants, particle and other Hierarchical identification [52] is another important
individuals in swarm intelligence optimization algorithm, branch of the system identification which is a new
and adjust individual’s state constantly according to the identification method based on identifying model
optimal strategy of swarm intelligence optimization decomposition and which is presented in order to solve the
algorithms, until the system error meets requirement system identification problem of large-scale systems with the
precision. Many scholars have proposed some new complex structure and multi-dimension. Its basic thought is
that the identification model is decomposed to make the

399
397
scale about identification small and to make the original [2] P. Eykhoff, “System Identification:Parameter and State Estimation,”
problem relatively simple and the calculation of John Wiley & Sons, 1974, pp. 555.
identification method becomes smaller. Hierarchical [3] L. Ljung, “Convergence analysis of parametric identification
methods,” IEEE Trans. on Automatic Control, vol. 23 no. 5, Oct.
identification based on the hierarchical identification theory 1978, pp. 770-783.
has become a new identification field it can be applied for [4] F. Ding, “System identification, Part A: Introduction to the
studying identification problems of all kinds of multivariable identification,” Journal of Nanjing University of Information Science
systems with the complex structure and nonlinear systems. and Technology: Natural Science Edition, vol. 3, no. 1, 2011, pp. 1-
These methods are as follows: hierarchical identification of 22, doi: 10.3969/j.issn.1674-7070.2011.01.001.
large-scale systems [53], gradient based iterative algorithms [5] Z. Pang, H. Cui, “MATLAB Simulation of System Identification and
for solving a class of matrix equations [54], hierarchical least Adaptive Control,” Beijing: National Defense Industry Press, 2009.
squares identification methods for multivariable systems [6] D. Liu, “System Identification Methods and Applications,” Beijing:
[55], hierarchical identification of state space models for National Defense Industry Press, 2010.
multivariable systems [56] and so on. [7] S. Li, W. Li, “Intelligent Control,” Harbin: Harbin Institute of
Technology Press, 2011.
IV. THE DEVELOPING TENDENCY AND PROSPECT OF [8] W. Wang, C. Li, J. Tian, “Identification of a class of nonlinear
systems using neural networks,” Journal of Qingdao University of
SYSTEM IDENTIFICATION TECHNOLOGY Science and Technology, vol. 26, no. 1, 2005, pp. 85-88, doi:
In recent years, as the constant development and 10.3969/j.issn.1672-6987.2005.01.022.
innovation of system identification technology, some new [9] J. Kang, W. Meng, “An improved method of system identification
methods of system identification are constantly emerging, based on neural network,” Computer and Digital Engineering, vol. 40,
no. 1, 2012, pp. 31-33, doi: 10.3969/j.issn.1672-9722.2012.01.012.
and obtain better identification effects in actual application.
[10] Y. Shen, X. Ju, C. Liu. “Application of second order diagonal
Although system identification technology has been recurrent neural network in nonlinear system identification,”
developed for several decades and has achieved many International Conference on Web Information Systems and Mining,
research results, there are still many problems needing to be 2010, pp. 420-424.
further discussed and studied. Some mainly aspects need to [11] N. Wang, Y. Tan, S. Liu, “A generalized online self-organizing fuzzy
be advised as follows: neural network for nonlinear dynamic system identification,”
• Reliance on system priori knowledge too much. Proceedings of the 30th Chinese Control Conference, July 2011, pp.
2879-2883.
• Model has poor universality.
[12] S. Ding, Q. Wu, Y. Yang, “Research on the application of RBF neural
• The identified model can not equivalent to measured network based on K-means clustering in system identification,”
system completely. Fourth International Workshop on Advanced Computational
Therefore, looking for a system identification method Intelligence, Oct. 2011, pp. 110-112, doi:
which needn’t priori knowledge, and whose versatility and 10.1109/IWACI.2011.6159984.
approximation capability is good remains an important [13] X. Xu, “Application of wavelet analysis in system identification,”
research direction in system identification realm. China New Technologies and Products, no. 2, 2010, pp. 19, doi:
10.3969/j.issn.1673-9957.2010.02.018.
V. CONCLUSION [14] L. Dai, “Research on nonlinear system identification based on
wavelet networks,” Fire Control and Command Control, vol. 31, no. 6,
After decades of development, system identification June 2006, pp. 44-47, doi: 10.3969/j.issn.1002-0640.2006.06.013.
theory has been studied and applied more and more deep and [15] Y. Xiao, “Based on wavelet transform algorithm for system
widely, especially it has achieved great success in identification,” University of Science and Technology Liaoning,
application of some research fields such as mechanism Anshan, 2008, chapter 5, doi: 10.7666/d.y1242605.
research and simulation of system, automatic control, [16] Y. Tan, X. Dang, F. Liang, “Dynamic wavelet neural network based
nonlinear dynamic system identification,” Journal of System
adaptive prediction and forecast, fault diagnosis and so on. Simulation, vol. 13, Aug. 2001, pp. 51-55, doi: 10.3969/j.issn.1004-
The author, from the different definitions of system 731X.2001.z1.017.
identification, reviews the traditional and modern methods of [17] D. Li, S. Huang, “Orthogonal least squares based wavelet networks in
system identification, and then analyzes the difficulties and system identification,” Control and Decision, vol. 18, no. 3, May
prospects of system identification, in hopes of bringing 2003, pp. 378-381.
benefits to related researchers and engineers. [18] S. Wilson, “Optimizing the performance of the wavelet-correlation
system identification method,” Conference Proceedings of IEEE
ACKNOWLEDGMENT Southeastcon, Mar. 2012, pp. 1-5, doi: 10.1109/SECon.2012.6196993.
[19] H. Shi, Y. Cai, Z. Qiu, “Improved system identification approach
The author would like to thank the financial supports using wavelet networks,” Journal of Shanghai University, vol. 9, no.
from National Natural Science Foundation of China under 2, April 2005, pp. 159-163.
Grant 61074090, Program for Liaoning Excellent Talents in [20] M. Davanipoor, M. Zekri, F. Sheikholeslam, “Fuzzy wavelet neural
University under Grant LR2011005, the AVIC Innovation networks with hybrid algorithm in nonlinear system identification,”
Funds under Grant cxy2011SH and Shenyang City Science IEEE International Conference on Computer Science and Automation
and Technology Project under Grant 131365. Engineering, no. 1, June 2011, pp. 153-156, doi:
10.1109/CSAE.2011.5953193.
REFERENCES [21] C. Sun, “Rule-base structure identification in an adaptive-network-
based fuzzy inference system,” IEEE Transactions on Fuzzy Systems,
[1] L. Zadeh, “From circuit theory to system theory,” Proc. of the IRE, vol. 2, no. 1, Feb. 1994, pp. 64-73, doi: 10.1109/91.273127.
vol. 50, no. 5, May 1962, pp. 856-865.

400
398
[22] S. Chiu, “Fuzzy model identification based on cluster estimation,” [41] H. Zhou, R. Wang, “Application of ABC algorithm in
Journal of Intelligent and Fuzzy Systems, vol. 2, no.3, 1994, pp. 267- identification and control of nonlinear system,” Chinese
278. Mechanical Engineering Society, vol. 23, no. 12, Aug. 2012, pp.
[23] S. Ertugrul, “Predictive modeling of human operators using para- 1446-1451.
metric and neuro-fuzzy models by means of computer-based i- [42] H. Zhang, “Improvement of artificial fish swarm algorithm and
dentification experiment,” Engineering Applications of Artificial application in parameter estimation for an induction motor’s
Intelligence, vol. 21, no. 2, June 2008, pp. 259-268, doi: model,” North China Electric Power University, Beijing, 2011,
10.1016/j.engappai.2007.03.007. chapter 4.
[24] H. Du, N. Zhang, “Application of evolving T-S fuzzy model to [43] F. Ding, “System identification. Part D: Auxiliary model
nonlinear system identification,” Applied Soft Computing, vol. 8, no. identification idea and methods,” Journal of Nanjing University of
1,May 2008, pp. 677-686, doi: 10.1016/j.asoc.2007.05.006. Information Science and Technology: Natural Science Edition, vol.
[25] C. Li, J. Zhou, X. Xiang, “T-S fuzzy model identification based on a 3, no. 4, January 2011, pp. 289-318, doi: 10.3969/j.issn.1674-
novel fuzzy c-regression model clustering algorithm,” Engineering 7070.2011.04.001.
Applications of Artificial Intelligence, vol. 22, no. 4, June 2009, pp. [44] F. Ding, T. Chen, “Combined parameter and output estimation of
646-653, doi: 10.1016/j.engappai.2009.02.003. dual-rate systems using an auxiliary model”, Automatica, vol. 40, no.
[26] A. Tewari, M. Macdonald, “Knowledge-based parameter identi- 10, Oct. 2004, pp. 1739-1748, doi: 10.1016/j.automatica.2004.05.001.
fication of TSK fuzzy models,” Applied Soft Computing, vol. 10, no. [45] F. Ding, J. Ding, “Least squares parameter estimation for systems
2, 2010, pp. 481-489, doi: 10.1016/j.asoc.2009.08.034. with irregularly missing data,” International Journal of Adaptive
[27] H. Anh, K. Ahn, “Identification of pneumatic artificial muscle Control and Signal Processing, vol. 24, no. 7, Jul. 2010, pp. 540-553,
manipulators by a MGA-based nonlinear NARX fuzzy model,” doi: 10.1002/acs.1141.
Mechatronics, vol. 19, no. 1, 2009, pp. 106-133, doi: [46] L. L. Han, J. Sheng, F. Ding. “Auxiliary model identification method
1016/j.mechatronics.2008.06.004. for multi-rate multi-input systems based on least squares,”
[28] C. Li, D. Sun, “The application of genetic algorithms to system Mathematical and Computer Modelling, vol. 50, no. 7/8, June. 2009,
identification,” Journal of Huazhong University of Science and pp. 1100-1106, doi: 10.1109/ACC.2009.5160447.
Technology, vol. 26, no. 7, Jul. 1998, pp. 57-58. [47] F. Ding, Y. Shi, T. Chen, “Auxiliary model based least squares
[29] X. Li, “An approach for system identification based on genetic identification methods for Hammerstein output error systems,”
algorithms and evolutionary programming,” Control and Decision, Systems and Control Letters, vol. 56, no. 5, May 2007, pp. 373-380,
vol. 11, no. 3, May 1996, pp. 404-407. doi: 10.1016/j.sysconle.2006.10.026.
[30] B. Jiang, B. Wang, “Parameter estimation of nonlinear system based [48] F. Ding, “System identification. Part F: Multi-innovation
on genetic algorithms,” Control Theory and Applications, vol. 17, no. identification theory and methods,” Journal of Nanjing University
1, 2000, pp. 150-152, doi: 10.3969/j.issn.1000-8152.2000.01.038. of Information Science and Technology: Natural Science Edition,
vol. 4, no. 1, May 2012, pp. 1-28, doi: 10.3969/j.issn.1674-
[31] S. Yang, H. Zhang, “MATLAB Technology Realization of Swarm 7070.2012.01.001.
Intelligence and Bionic Calculation,” Beijing: Publishing House of
Electronics Industry, 2012. [49] F. Ding, D. Y. Xiao,T. DING, “Multi-innovation stochastic
gradient identification methods,” Control Theory and Application,
[32] A. Colorni, M. Dorigo, V. Maniezzo, “Distributed optimization by
vol. 20, no. 6, Dec. 2003, pp. 870-874, doi: 10.3969/j.issn.1000-
ant colonies,” Proceedings of the first European Conf on Artificial 8152.2003.06.011.
Life, 1991, pp. 134-142.
[50] F. Ding, H. B. Chen, M. Li, “Multi-innovation least squares
[33] R. Eberhart, J. Kennedy, “A new optimizer using particle swarm
identification methods based on the auxiliary model for MISO
theory,” Proceedings of the Sixth International symposium on
systems,” Applied Mathematics and Computation, vol. 187, no. 2,
Micro Machine and Human Science, Oct. 1995, pp. 39-43, doi: April 2007, pp. 658-668, doi: 10.1016/j.amc.2006.08.090.
10.1109/MHS.1995.494215.
[51] Y. Y. Liu, D. R. Shen, Y. J. Chen, “Multi-innovation stochastic
[34] K. Passino. “Biomimicry of bacterial foraging for distributed gradient identification algorithm based on feed-forward neural
optimization and control,” IEEE Control Systems Magazine, vol. networks,” Journal of Harbin University of Commerce: Natural
22, no. 3, June 2002, pp. 52-67, doi: 10.1109/MCS.2002.1004010.
Sciences Edition, vol. 22, no. 2, May 2006, pp. 83-86, doi:
[35] B. Basturk, D. Karaboga, “A powerful and efficient algorithm for 10.3969/j.issn.1672-0946.2006.02.024.
numerical function optimization: artificial bee colony (ABC) [52] F. Ding, “System identification. Part G: Hierarchical identification
algorithm,” Journal of Global Optimization, vol. 39, 2007, pp. principle and methods,” Journal of Nanjing University of
459-471.
Information Science and Technology: Natural Science Edition, vol.
[36] X. Li, Z. Shao, J. Qian, “An optimizing method based on 4, no. 2, Jul. 2012, pp. 97-124, doi: 10.3969/j.issn.1674-
autonomous animats: Fish-swarm Algorithm,” Systems 7070.2012.02.001.
engineering theory and Practice, vol. 22, no. 3, Nov. 2002, pp. 32- [53] F. Ding, J. B. Yang, “Hierarchical identification of large scale
38. systems,” Acta Automatica Sinica, vol. 25, no. 5, 1999, pp. 647-654.
[37] L. Wang, Q. Wu, “Ant system algorithm based system [54] F. Ding, T. Chen, “Gradient based iterative algorithms for solving a
identification,” ACTA AUTOMATIC ASINICA, vol. 29, no. 1, class of matrix equations,” IEEE Transactions on Automatic Control,
Jan. 2003, pp. 102-109.
vol. 50, no. 8, Aug. 2005, pp. 1216-1221, doi:
[38] S. Hou, “Study on parameter identification of linear system based 10.1109/TAC.2005.852558.
on self-adaptation ant colony algorithm,” Journal of Hunan
[55] F. Ding, T. Chen, “Hierarchical least squares identification methods
Institute of Engineering, vol. 20, no. 3, Sept. 2010, pp. 55-58, doi:
for multivariable systems,” IEEE Transactions on Automatic Control,
10.3969/j.issn.1671-119X.2010.03.017.
vol. 50, no. 3, Mar. 2005, pp. 397-402, doi:
[39] X. Xu, F. Qian, D. Liu, “Method of system identification based on 10.1109/TAC.2005.843856.
PSO algorithm,” Journal of System Simulation, vol. 43, no. 2, Jul.
[56] F. Ding, D. Y. Xiao, “Hierarchical identification of state space
2009, pp. 116-120.
models for multivariable systems,” Control and Decision, vol. 20, no.
[40] J. Jin, “A novel method for parameter identification of process 8, Sep. 2005 848-853, 859.
model,” Control Theory and Applications, vol. 31, no. 1, April
2012, pp. 4-7, doi: 10.3969/j.issn.1003-7241.2012.01.003.

401
399

You might also like