Dynamic retail market tariff design for an electricity aggregator using reinforcement learning

Abstract

The role of retailers, as energy providers for end-users, in restructured retail electricity markets becomes substantial. The increasing share of distributed energy resources and electrification in different sectors bring several challenges to retailers. Among these challenges, procuring electricity and maintaining system reliability during peak times induce high costs to retailers. Therefore, they need to accurately predict customers’ demand to participate in the wholesale market and develop proper tariff mechanisms considering other retailers’ behavior to maximize their profit. This paper develops the design of an autonomous retailer in which a Sequence-to-Sequence (Seq2Seq) algorithm is employed to predict customers’ net demand. Furthermore, using Reinforcement Learning (RL), the proposed retailer designs tariff mechanisms based on other retailers’ behavior and customers’ load profiles. The proposed design of the retailer is evaluated on a retail market simulation platform called Power TAC, in which autonomous retailers compete in retail, wholesale, and balancing markets to maximize their profits. The results show the accuracy of the proposed load prediction method compared with other methods and successful profit growth with a drop in fixed costs and balancing costs.