Neuro-Fuzzy Adaptation of Synergistic Control System of Multi-Mode Objects

Abstract

The article considers the issues of creation of high-efficiency
algorithms for control of technological facilities, functioning in conditions of
uncertainty. The algorithm of neuro-fuzzy adaptation of fuzzy-logical PIDregulator is proposed, which allows ensuring high speed of the synergistic control
system of multi-mode objects due to reduction of the number of iterations during
training, that is, the number of neural network training epochs. A fast fuzzy-logic
output algorithm has been developed to eliminate empty solutions and zero
sections in terms that describe input and output fuzzy variables. A structural
diagram of synergistic control systems of multi-mode technological processes has
been developed, which includes an adaptation unit, which allows correcting not
only the parameters, but also the structures of the control stages. Proposed neurofuzzy adaptation in control tasks of technological process equipment allows
accelerating process of system training, due to use of high-speed algorithm of
fuzzy-logical output, to reduce error of results of training of neuro-fuzzy network
from 8 to 1%. An algorithm for training the neural network was developed, based
on the use of the soft computation method, using the area difference of the
belonging function. Based on the simulation experiment, it is determined that the
accuracy of training in the existing methods is about 5%, therefore one parameter
must be corrected, by change in the third or fourth layer of the neuro-fuzzy
network fuzzy-logical operations, that is, instead of min, use the max operation,
which allows to obtain a given result in fewer iterations than in the known
methods.