Implementing Reinforcement Learning for Tackling Smart Grid Pricing Problem

In the smart grid system, dynamic pricing can be an effective instrument for the service provider. In practice, however, the paucity of information regarding customers’ time-varying load demand and energy consumption patterns, as well as the volatility of wholesale electricity prices, make the implementation of dynamic pricing extremely difficult. In this paper, we examine a dynamic pricing problem in the smart grid system where the retail electricity price is determined by the service provider. To surmount the obstacles associated with implementing dynamic pricing, we devise an algorithm for reinforcement learning. The main challenge tackled by this paper is obtaining a pragmatic solution for satisfying the conflicting criteria—high profit of power utilities and low energy cost for consumers. Results from numerical simulations indicate that the proposed reinforcement learning algorithm can function effectively without prior knowledge of the system dynamics. During peak hours, the service provider profit was as high as approximately 6297 units—which is an impressive gain.