Smart Homes, Smarter Savings: Energy Trading with Deep Reinforcement Learning

Over the last decade, many countries have introduced dynamic energy pricing to curb peak electricity demand of consumers. The dynamic pricing provides smart homes equipped with energy generation units and Energy Storage Systems (ESS) the opportunity to smartly utilise these resources to both balance demand and reduce energy costs. In this paper, we consider an individual smart home energy management system that can trade energy with the electric grid and does not intrude in the energy consumption habits of the homeowner. This particular aspect has hardly been studied so far, as most research focuses on the planning of household appliance scheduling and omit the possibility of energy trading. To this end, we have developed a Deep Reinforcement Learning (DRL) solution that significantly reduces energy costs in the smart home and in many cases going as far as producing profits for the homeowner. Additionally, we run simulations in various environments, in particular we use two dynamic pricing models and investigate the optimality of the solution considering different ESS charging and discharging losses.