НЕЙРОСЕТЕВОЙ ПОДХОД К ПОСТРОЕНИЮ МАРШРУТА В АВТОМАТИЗИРОВАННОЙ СИСТЕМЕ УПРАВЛЕНИЯ СПЕЦИАЛЬНОГО НАЗНАЧЕНИЯ

Abstract

This work consists of the introduction, statement of the problem, descriptions of the method used and initial data for modeling, analysis of the results of the implementation of the methods, conclusions. The introduction substantiates the relevance of the problem consisting of the need to build a route in a telecommunications network in several metrics. The analysis of traditional methods is carried out and an alternative method is proposed in the form of an artificial Hopfield neural network. The goal of the presented work is formulated as follow in implement a neural network algorithm for the constructing a route in an automated control systems to assess the possibilities of its practical application in real systems. The statement of the problem uses general terms to mathematically formalize the task of the route constructing; concepts such as an undirected graph, an arc of a graph, and the cost of an arc are introduced. As a result, the main routing problem is formulated as the task of the finding the path from the sender node to the minimum-cost recipient node, where the minimum cost is the sum of the costs of the arcs that make up the constructed path. When describing the used method of the route constructing in an automated control systems system, the basic concepts (architecture and characteristics) of the approach under consideration are given. The sequence of the modification of the Lyapunov energy function identifying the state of the Hopfield neural network is given, it’s main disadvantages are noted and the final form of the energy function in the form of five terms is given. In particular, the first term not only minimizes the total cost of the channel in the route, taking into account the cost of existing channels, but also takes into account all the cost values around the considered node. The second term prevents the inclusion of no existing lines in the selected route option. The third term is equal to zero if the number of incoming communication directions is equal to the number of arcs in the outgoing direction, while the fourth term ensures the ability of the neural network to converge to the actual route. The fifth term prevents the appearance of loops in the route. The article describes the dependence that determines the dynamics of changes in the inputs of the Hopfield neural network. Based on the above dependencies, a block diagram of the algorithm for the functioning of an artificial neural network for the formation of a route in an automated control systems is developed. Then the initial data in the form of a network structure, the values of the coefficients of the energy function are provided. The conclusions of the work formulate the main tasks that are to be solved to use the neural network approach to construct routes in real networks as well as proposals for the confirming of the consistency of the considered method from the viewpoints of the system’s integral indicators.