Page 7 - Neural Network Modeling and Identification of Dynamical Systems

Preface

One of the key elements of the development tems operating under significant and heteroge-
process for new engineering systems is the for- neous uncertainties. The presence of models of
mation of mathematical and computer models this kind opens up new opportunities for solv-
that provide solutions to the problems of creat- ing problems of control of complex dynamic sys-
ing and using appropriate technical systems. As tems.
the complexity of the created systems grows, so The introductory chapter substantiates the
do the requirements for their models, as well as need for a new approach to the formation of
for the funds raised for the development of these mathematical and computer models of con-
models. trolled dynamical systems operating under con-
At present, the possibilities of mathematical ditions of multiple and heterogeneous uncer-
and computer modeling are lagging behind the tainties. We demonstrate that the control prob-
needs of a number of engineering fields, such lems for dynamical systems under uncertainty
as aerospace technology, robotics, and control of conditions cause a need to give the property
complex production processes. Characteristic of of adaptability to the dynamical system model.
technical systems from these areas is the high The traditional approach to mathematical and
level of complexity of the objects and processes computer modeling of the dynamical systems
being modeled, their multidimensionality, non- does not satisfy this requirement. We can over-
linearity, and nonstationarity, and the diversity come this difficulty by applying techniques of
and complexity of the functions implemented neural network modeling. However, traditional
by the object being modeled. Solving the prob- ANN models, which belong to the black box
lems of modeling for objects of this kind is sig- class, do not allow for a complete solution to the
nificantly complicated by the fact that the corre- task. This circumstance makes it necessary to ex-
sponding models have to be formed in the pres- pand the black box–type neural network models
ence of multiple and heterogeneous uncertain- to the gray box class.
ties, such as incomplete and inaccurate knowl- Chapter 1 deals with the modeling of con-
edge of the characteristics and properties of the trolled motion of nonlinear systems. It covers
object being simulated, as well as the conditions topics such as a dynamical system as an object
in which the object will operate. In addition, of study, the formalization of the concept of the
the properties of the simulated object may un- dynamical system, and the behavior and activi-
dergo changes, including sharp and significant ties of such systems. The concept of adaptability
changes immediately during operation, for ex- is discussed, one of the most important concepts
ample, due to equipment failures and/or dam- for the advanced dynamical systems being cre-
age of its structure. In this case, an object model ated, in particular for robotic aircraft of various
that was previously formed on the basis of its classes. An approach is formulated to solve the
nominal (“intact”) state becomes inadequate. If problem of dynamical system modeling, a gen-
this model is used, for example, in an object con- eral scheme of the modeling process is given,
trol system, a critical situation arises. and the main problems that need to be solved
In this regard, it is reasonable to search for when forming the dynamical system model are
new tools for modeling nonlinear controlled sys- identified.

ix