The problems of synthesis of a model of dynamic neuron with state memory (DNSM) are considered in the paper. The introduction of a special additional parameter into the model of a neuron, defined as a state parameter, is substantiated. It is indicated that the parameter of the state of the neuron has the ability to vary with time depending on the nature of the information processes that occur in neighboring neurons of the network. This parameter in a certain way accumulates information about the history of the behavior of the neuron in accordance with the entered formal descriptions. The concept of a "strained neuron" is introduced, taking into account the above. This concept characterizes the degree of influence of a given neuron on the neurons surrounding it. On the effects of time-varying parameters of the state of neurons, it is proposed to implement the process of self-evolution of the network directly during its operation. A variant of the analysis of the structure of the neural network, created on the basis of the proposed model DNSM. The topological representation of a neural network in the form of a graph model allows formalizing the interaction of neurons in a network with each other, both in time and in space. For this, the concept of k-space is introduced, which determines the degree of proximity of neurons to each other. The degree of proximity of neurons allows one to formalize, in the form of mathematical relationships, the procedure for the exchange of information between neighboring neurons in a network. Mathematical relationships that formalize these processes are given. A variant of the structure of the hardware design of DNSM, focused on implementation using FPGA technology, is proposed.

Keywords: dynamic neuron with state memory, connectionist model, self-evolutionary mechanism

The article considers a number of proposals aimed at improving the efficiency of the design, training and operation of neural network models within the framework of control systems for complex objects. This includes: organization of processing of networks of large dimenstionality; development of neuroevolutionary procedures; providing opportunities for the formation of neural network models in a single cycle of designing automated systems. The hierarchical construction of network structures is considered as an apparatus that provids work with network models of high dimensionality. At the same time, unlike the traditional approach associated with cutting the graph of an already synthesized model, the article proposes procedures for synthesizing a generalized model from structural elements. The formal descriptions of structural synthesis introduced in the article can serve as a basis for organizing software and hardware interfaces of the forming models. The formation of such neural network structures within control systems is associated with the proposed and described concept of neural network regulators as the basic elements of building intelligent automated systems. The article also examines and describes as a neuroelement a model of a dynamic neuron with a state memory, allowing to implement the mechanism of self-evolutionary development of the network directly in the course of its operation.

Keywords: control / management system, dynamic neuron, neural network model, hierarchical structure, self-evolution mechanism