Minimizing Grid Interaction of Solar Generation and DHW Loads in nZEBs Using Model-Free Reinforcement Learning

Abstract

This study applies model-free reinforcement learning (RL) on a case study based in Utrecht province in the Netherlands to optimize for on-site renewable energy. This aims at reducing the interaction of net zero-energy buildings with the grid as a result of an increase of heat pump installations and renewable energy systems (RES) integration. It is believed that this will become increasingly more important since the regulations regarding 2020 and beyond ascribe significant increase in energy efficiency of the built environment. On-site RES self-consumption is therefore a central lead in this research. The project data comprise air source heat pump and solar energy data of 6 different households for the months June to November 2016. The RL algorithm is applied to the different data sets to derive an optimized individual and generalized control strategy. Simulations were carried on, to acquire the resulting energy consumption, self-consumption, and self-sufficiency. The results show an increase of individual self-consumption between 17% and 348% and self-sufficiency between 18% and 72%. This results in an additional monetary benefit for the occupants based on the transition proposals of 2020 for the renewable energy generation net-metering abolishment in the Netherlands. Furthermore, reducing the grid interaction implies benefits for the grid operators in terms of investments required for grid reinforcement.