Page 113 - Neural Network Modeling and Identification of Dynamical Systems

102 3. NEURAL NETWORK BLACK BOX APPROACH TO THE MODELING AND CONTROL OF DYNAMICAL SYSTEMS

Chapter 5, and we present examples of their advanced aircraft technology remains a chal-
use in Chapter 6. lenging task so far. First of all, this is because a
modern multipurpose highly maneuverable air-
craft should operate under a wide range of flight
3.4 ANN-BASED CONTROL OF conditions, masses, and inertial characteristics,
DYNAMICAL SYSTEMS with a significant nonlinearity of the aerody-
namic characteristics and the dynamic behavior
of aircraft.
The development of complex controlled sys- For this reason, it seems relevant to try to ex-
tems raises some problems that we cannot solve pand the range of methods and tools tradition-
through traditional control theory alone. These ally used to solve aircraft control problems, bas-
tasks associate mainly with such uncertainty in
ing these methods on the approaches offered by
the conditions of the system operation, which intelligent control. One such approach is based
requires the implementation of decision-making on the use of ANNs.
procedures that are characteristic of a person
with the ability to reason heuristically, to learn,
to accumulate experience. The need for train- 3.4.1 Adjustment of Dynamic Properties
ing arises when the complexity of the problem of a Controlled Object Using
being solved or the level of uncertainty under Artificial Neural Networks
its conditions does not allow for obtaining the
In this section, an attempt is made to show
required solutions in advance. Training in such that using ANN technology we can solve the
cases makes it possible to accumulate informa- problem of appropriate representation (approx-
tion during the operation of the system and use
imation) of a nonlinear model of aircraft motion
it to develop solutions that meet the current sit-
with high efficiency. Then, using such approx-
uation dynamically. We call systems that imple-
imation, we can synthesize a neural controller
ment such functions intelligent control systems.
that solves the problem of adjusting the dynamic
In recent years, intelligent control, which
properties of some controlled object (aircraft).
studies methods and tools for constructing and First, we state the problem of adjusting the
implementing intelligent control systems, is an dynamic properties of a controlled object (plant),
actively developing field of interdisciplinary re- based on an indirect assessment of these prop-
search based on the ideas, techniques, and tools erties using the reference model. It is proposed
of traditional control theory, artificial intelli- to solve this problem by varying the values of
gence, fuzzy logic, ANNs, genetic algorithms, the parameters of the controller, producing ad-
and other search and optimization algorithms. justive actions on the plant.
Complex aerospace systems, in particular, air- Then a structural diagram of varying the pa-
craft, entirely belong to these complex controlled rameters of the adjusting (correcting) controller
systems. In the theory of automatic control and is proposed using a reference model of the be-
its aerospace applications, significant progress havior of the plant. We show that we have to re-
has been made in the last few decades. In partic- place the traditional model of aircraft motion in
ular, considerable progress has been made in the the form of a system of nonlinear ODEs with an-
field of computing facilities, which allows carry- other model that would have substantially less
ing out a significant amount of calculations on computational complexity. We need this replace-
board of the aircraft. ment to ensure the efficient operation of the pro-
However, despite all these successes, the syn- posed control scheme. As such an alternative
thesis of control laws adequate to modern and model, it is suggested to use an ANN.