Method of synthesis and automatic adaptation of the architecture of a hierarchical multi-agent system

Background. The architecture of a multi-agent system (MAS) is a set of connections between agents, their roles, and rules governing their behavior. The effectiveness of a MAS largely depends on the choice of architecture. During operation, situations may arise that require prompt modification of the MAS architecture (changes in environmental parameters, malfunctions, and failures of agents). The variability of agent operating conditions (various environmental options, types of tasks) requires greater flexibility in configuring the MAS architecture, which existing solutions cannot provide. The object of the study is multi-agent systems. The subject of the study is methods for forming the MAS architecture. The aim of the work is to develop a method for synthesizing and automatically adapting the architecture of a hierarchical MAS. Materials and methods. Reinforcement learning paradigm methods, genetic algorithm. Results. As a result, a method for synthesizing and automatically adapting the architecture of a hierarchical MAS was developed, characterized by automatic modification of a set of agent connections during the operation of the MAS (as well as basic rules that determine the conditions for the emergence of connections), the ability to determine the optimal parameters of the MAS for a specific environment using a genetic algorithm, as well as the ability to model several types of MAS architectures. Conclusions. The developed method can find its practical application in the implementation of the following tasks: inspection (or patrolling) of infrastructure facilities by mobile robots; implementation of artificial intelligence in computer games.