Adaptive Multiagent Model Based on Reinforcement Learning for Distributed Generation Systems

Distributed generation have been widely spread in the last decades raising a lot of questions regarding the safe and high-quality operation of the power systems. The investigation of these questions requires a proper model considering the different technical, economical and legal aspects. The goal of our research was to develop a multiagent system where rational agents control each distributed generation unit. Based on intelligent agent-program the agents are able to optimize their operations taking several viewpoints into account, like fulfilling the contractual obligations, considering the technical constraints and maximizing the realized profit in a continuously varying market environment. This paper describes a simple reinforcement learning method resulting in an adaptive agent-program. The agents are informed about their realized profits and they apply this information to evaluate their former decisions and to adjust the parameters of their agent-program. The verification of the model proved that the developed agent-program provides acceptable results compared to the real productions.