ASSOCIATING AUTOMAT FOR TECHNOLOGICAL PROCESSES ADAPTIVE CONTROL ON BASED OF NEURAL NETWORKS

Author Name(s): A.V. Daneev, V.N. Sizykh
Author Email:

Abstract

The problem of physical object neural control with parametric uncertain is considered. The architecture to organizations of the technological process through parallel interaction of technological modules (agents) similar one-layer neural-fuzzy control is offered. On based of the Lyapunov’s direct research method of absolute stability and speed gradient method both a nonlinear algorithm of the operative education and control is designed. The developed algorithm of adaptation allows accelerating a convergence neural-fuzzy control through the neural network training in real time, carrying out preliminary procedures training and to apply an effective ways of parametrical their initialization. Through the additional feedback on control and the sliding modes organization about artificial neuron special points both the invariance property to external, parametrical and dynamic object noise is provided. At the organization similar neural network structures the Gurvitz’s state matrixes condition for system linear part can’t be carried out. Besides, area of admissible values of regulator’s parameter is all set of real numbers. Software and hardware implementation of decision-making process an associating automat of adaptive control is expected to implement in informational real time systems on based of the Iu. F. Muhopad`s structural model.

Introduction

The complex control system of difficult technical system consists of object of control and the actuation device. In mechanical engineering object of control is the technological process (TP). TP is the part of production containing actions for change and the subsequent definition of a condition of a subject of production [1]. TP as object of management is the difficult dynamic system including a number of subsystems operations (technological modules (TM) [2]), each of which separately too can be object of management. It is noted [2] that the main complexity of management of TP consists in insufficient degree of its study, in limitation of volume and in incompleteness of measurements. From positions of cybernetic physics [3] the technical system is a physical system with entrances and exits which functioning happens without participation of the person at the expense of the organization of feedback between her entrances and exits.

Conclusion

Thus, in article on the basis of the organization of parametric control for a method of a high-speed gradient and the theory of systems with variable structure the effective algorithm of adaptation allowing is developed:

1) to accelerate convergence neuro-indistinct control due to training of a neural network in real time;

2) to apply hybrid networks on which artificial neural networks communicate with structures of the adaptive control on the basis of traditional technologies;

3) to realize preliminary training and to apply effective procedures of initialization of parameters of a neural network.

For implementation of an algorithm of adaptation on the basis of a one-layer neural network there are enough knowledge of an order of the differential equations describing a physical entity with steady structure of correlations between its elements. Unlike traditional adaptive approach when the area of steady solutions of the linear part of system () is defined by a condition , in case of the organization of neurosimilar structures area of admissible parameter values L of the regulator is all set of the real numbers .

1,725 total views, no views today

Download PDF File

About the author: admin