Intelligent Agent Strategies for Residential Customers in Local Electricity Markets

Abstract

The energy transition from a formerly centralized, fossil-fuel based system towards a sustainable system based on a large share of renewable generation calls for a decentralization and regionalization of the electricity system. Local electricity markets (LEMs), on which prosumers and consumers can trade locally produced electricity, meet these requirements and simultaneously enable formerly excluded residential customers to actively take part in the electricity market. However, trading can be complex and time intensive. Therefore, it should be automated. We provide an analysis of intelligent learning strategies for agents of residential electricity customers in LEMs. To this end, we conduct a multi-agent-based simulation of a LEM with a merit order market design based on the current German electricity spot market. LEM agents maximize their individual utility via reinforcement learning. We expand existing approaches of reinforcement learning with generation and storage states as well as time-dependent learning. The evaluation of the strategies is based on the agents' and community electricity storage's revenues, costs, and consumption of local electricity. The results show that for fixed sell prices, time-dependent reinforcement learning of buy bids is the best strategy. It facilitates a market self-consumption of 54 %. For learning buy and sell prices, traditional reinforcement learning with generation states is the dominant strategy.