Adapting the power grid to renewable energy source is a significant challenge for the upcoming years. One way to shift the energy demand to the peak times of renewable energy generation can be enabled through consumer-side control systems.
We aim to develop such a control system using reinforcement learning. For this purpose, the framework and data of the NeurIPS 2022 Citylearn Challenge will be used. Unlike previous models in CityLearn, we choose transformers as our approach due to impressive success in many areas of machine learning. In this work, we will therefore specifically investigate the Decision Transformer model presented by Chen et al. for its applicability as an energy control system.
Reference repository http://gitlab.aicrowd.com/adrien_forbu/neurips-2022-citylearn-challenge.git