Cooperative behavior is understood as a community of agents who decide to cooperate to reduce the average weighted fines to solve a task or achieve a certain goal, in our case, synchronize the lightning. The problem of forming cooperative behavior is intensively investigated in modern scientific literature on the use of multi-agent systems, for example, for distance learning, management of organizational systems, the construction of various virtual organizations and communities, management of

more »

... stributed computing, management of public institutions and socio-economic processes, and others. In this paper an actual theme of optimal policies in games with local interaction is considered, the stimulating training of multiagent systems in gaming is considered. The purpose of this work is to consider the method of constructing a system with local interaction of agents based on the task of "synchronization" with the help of the Markov model of stochastic game. The research method is a computer program for modeling a task using the Q-method of training. Formation of coalitions in multiagent systems is formulated as a competitive or cooperative task of assigning an object to one of the clusters. The problem of solving such problems is the theory of games, and in the conditions of uncertainty the theory of stochastic games. In this regard, from the scientific and practical point of view, the use of stochastic game methods for the formation of coalitions under conditions of incomplete information is relevant. The decision of the stochastic game is to find policies for agents that maximize their winnings to provide a certain collective balance of interests for all players. To find optimal players' policies under uncertainty, we will use the method of stimulating learning. The result is a developed game model that provides a dynamic MAS self-organization, which manifests itself in the rhythmic change of pure agent policies that mimic the light effects of the colony of fireflies. A characteristic feature of the considered game self-organization is locally defined information about the policies of the behavior of neighboring agents, which as a result of learning leads to global coordination. policy of all agents. 57 # 24 (2018) ПРОБЛЕМИ ІНФОРМАЦІЙНИХ ТЕХНОЛОГІЙ The scientific novelty of the obtained results consists in the development of a gaming model, the effectiveness of the game self-organization of the MAS policies for solving the decision-making problem in systems with co-operative behavior of agents under uncertainty has been determined. The repetition of the values of the characteristics of the game in various experiments with unique sequences of random variables confirms the reliability of the results. Results can be used in practice to model the dynamics of social processes, management of social Internet services in the Internet and others.