Predicting future low voltage loads: extension of a method for optimised energetic building retrofitting using deep reinforcement learning

Increasing the energy efficiency of residential buildings is a major component of the energy transition in Germany. With the growing adoption of decentralised renewables in retrofitting efforts, there is an emergence of novel grid load scenarios characterised by enhanced uncertainty at the low voltage level. Nevertheless, for ensuring a stable grid operation, research is required to assess the implications of the energy transition. This paper presents a deep reinforcement learning framework for optimised energetic building retrofitting to simulate future low voltage load profiles. Within this framework, several categories of discrete building retrofit options are considered and an optimal policy for improving economic building efficiency is approximated. The outcomes indicate substantial efficiency improvements, with the outdoor temperature patterns of the respective building location playing a key role in determining the optimal actions. From the resulting retrofit concepts, high resolution load profiles can be generated and a prediction tool for estimating future grid scenarios can be derived.