Abstract:
The architecture of a multi-agent system (MAS) is a set of roles and relationships of agents, as well as a set of basic rules that determine their genesis. The objective of this work is to develop a method with the following char-acteristics: the ability to modify the basic rules that determine the conditions for establishing and breaking vertical and horizontal links between agents; the ability to automatically adapt the architecture (in the process of solving the problem) to changes in the values of environmental indicators, as well as in the quantitative and qualitative composition of the set of agents; the ability to model more than three MAS architectures. The developed method of initial formation and subsequent adaptation of the architecture of a multi-agent system fully satisfies the specified characteristics. The optimal set of basic parameters of a MAS is determined using a “genetic algorithm” by training on a virtual model of the environment. It is possible to model the following types of MAS architectures: “coalition”, “team”, “federation”, “congregation”. The conducted experimental evaluation confirmed the efficiency of the developed method in terms of the “total time to complete the assigned task” indicator in comparison with the following approaches: application of the “genetic algorithm” to the “chromosome” containing a set of vertical connections between agents (the architecture of the MAS can be changed only at a new iteration); the set of vertical connections is fixed and remains unchanged at each new iteration. The advantage of the proposed approach in comparison with its analogues is the ability to automatically adapt the architecture of the MAS in the process of solving the assigned task (and not at the beginning of a new iteration), adapting to variations in the set of agents or to new values of the environmental parameters.
Fulltext:
PDF file (5562 kB)