This paper examines the pressing issue of designing multi-agent systems (MAS) that require high adaptability in the face of dynamically changing environmental parameters, task conditions, and the internal structure of the MAS. A comprehensive method is proposed that combines the processes of synthesizing the initial architecture and subsequently automatically adapting both the MAS architecture, which defines the basic rules of its operation, and the organizational structure, which represents a hierarchical attributed directed graph in which the vertices are agents, and the edges correspond to the relationships between them. The method is based on the author's "zoosocial model" for representing the MAS architecture. It also utilizes an artificial neural network to predict optimal actions for modifying the MAS architecture in the event of changes in environmental parameters and task conditions. The REINFORCE algorithm is used for training the ANS, and a genetic algorithm is used to generate the training sample.

Keywords: multi-agent system, architecture, organizational structure, automatic adaptation, synthesis, genetic algorithm, reinforcement learning

The architecture of a multi-agent system defines the basic principles of its formation and operation, including the format of the organizational structure representing a graph in which agents act as vertices, and the links between them are designated by edges. A common drawback of existing approaches to representing the architectures of multi-agent systems is the support of no more than two types of organizational structures, among which the optimal one for the given environmental parameters may be absent. This paper proposes a method for representing the architecture of a multi-agent system, implemented by borrowing the mechanisms of living nature, namely the principles of organizing animal communities. The proposed approach allows modeling organizational structures of the following types: "coalition", "team", "hierarchical structure", "federation", "congregation". To determine the optimal architecture of a multi-agent system, optimal for specific environmental conditions, it is possible to use a "genetic algorithm".

Keywords: multi-agent system, architecture, agent, organizational structure, optimization