×

Nonlinear system identification. From classical approaches to neural networks and fuzzy models. (English) Zbl 0963.93001

Berlin: Springer. xvii, 785 p. (2001).
This volume in four parts presents several practical methods for system identification with an emphasis on the nonlinear case.
Part one is devoted to computational methods used in identification (linear optimization using the least squares approach), nonlinear local (mainly gradient based algorithms) or global (simulated annealing, genetic algorithms, branch-and-bound algorithms) identification, unsupervised learning (clustering, neural nets, fuzzy methods). The last chapter deals with model complexity optimization.
Part two presents various static models: classical (linear, polynomial, look-up tables), neural networks, fuzzy models, neuro-fuzzy models. The end of this part explores in more detail “local linear neuro-fuzzy models”.
Part three presents various dynamic models: linear and nonlinear, polynomial, neural networks, fuzzy or neuro-fuzzy models and focuses finally on the “local linear neuro-fuzzy models”.
The last part shows several applications: the combustion engine (which has received increased attention in the automotive industry recently due to new regulations in the European community), heat exchangers. A chapter introduces predictive control, fault detection and diagnosis.
Two appendices deal with matrix algebra and statistical methods. There are 414 references, and a medium sized index.
To quote the author: “The emphasis…is on an intuitive understanding of the subject and the practical application of the discussed techniques. It is not written in a theorem/proof style; rather the mathematics is kept to a minimum and the pursued ideas are illustrated by numerous figures, examples and real-world applications”. Later: “Unnecessary formalism and equations are avoided and the basic principles are illustrated with many figures and examples”. So, the author prefers figures over equations and seems to believe that this is a justification for advocating a better understanding of the subject. He also states: “Implementation details are skipped whenever they are not necessary to motivate and understand the approach”. Consequently, this book between mathematics and implementation is “written from an engineering perspective,” and “tailored for engineers in industry” in the hope that they will understand the topic better. Interestingly, the phenomenon of bifurcation (where the nonlinear dynamics changes when parameters vary) and the impact on nonlinear identification are not mentioned. Chaos is also absent and some work is being done on the identification of chaotic dynamics using possibly statistical methods. So this voluminous book cannot pretend to be complete.

MSC:

93-01 Introductory exposition (textbooks, tutorial papers, etc.) pertaining to systems and control theory
93-02 Research exposition (monographs, survey articles) pertaining to systems and control theory
93B30 System identification
93C10 Nonlinear systems in control theory
93C42 Fuzzy control/observation systems
92B20 Neural networks for/in biological studies, artificial life and related topics
93C15 Control/observation systems governed by ordinary differential equations
PDFBibTeX XMLCite