Abstract
In this paper, we describe an adaptive trading agent which can sell and buy electric power effectively in a locally produced and consumed electric energy network, ECONET (Electric Power Cluster Oriented Network). The trading agents manage the amount of electric power generated by solar panels or other renewable energies and stored in a storage battery in a minimal cluster. The agent learns a trading strategy by maximizing future cumulative reward based on reinforcement learning method. Especially, we build autonomous trading agents based on the natural actor-critic method, which is a type of natural policy gradient methods. Several experiments show that the adaptive trading agents can reduce useless energy consumption and a deficiency in most cases.
