Buildings play a crucial role in the global initiative to combat climate change, being responsible for a significant 30% of greenhouse gas emissions. At the same time, there is a growing recognition of the active role that buildings can play in supporting the energy system by providing flexibility to the electrical grid.
This project aims to develop a building-level control system using reinforcement learning techniques. Building on our previous experience with the NeurIPS Citylearn Challenge 2022, our goal is to refine and improve our control system, now incorporating considerations for occupant comfort and power outage resilience introduced in the Citylearn Challenge 2023.
Unlike previous models in CityLearn, we choose transformers as our approach due to their 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. leveraging its potential as powerful tool for enhancing our energy control system.
