MERLIN published in Applied Energy

We are thrilled to announce the publication of MERLIN, a significant milestone in the Intelligent Environments Laboratory’s journey to address real-world challenges of adopting Reinforcement Learning Control (RLC) for grid-interactive buildings and demand side energy management. 🏢🔌

MERLIN showcases the integration of model-free, offline, online, and transfer learning for scheduling battery energy storage systems in residential homes. We’ve tested it on 17 single-family prosumer homes in the Sierra Crest Zero Net Energy community in Fontana, CA. 🏘️🌞

MERLIN maintains privacy, adapts to individual load profiles, and doesn’t require system identification. Plus, it can be jump-started using a Time-of-Use (TOU) based scheduler, significantly reducing data requirements. We’ve implemented MERLIN in our CityLearn environment ( 🌐💻

Here’s what MERLIN achieves: it learns to shift and shed the load of individual homes based on pricing and carbon content signals from the grid. The result? A flattened aggregated load across the homes, reduced individual average daily peaks, and decreased electricity costs for homeowners. But that’s not all - it also leads to a significant reduction in carbon emissions and grid ramping, all of which significantly outperform the Time-of-Use (TOU) scheduler, and increase the value of home battery energy storage systems. 🏠💡🌍

Kudos to IEL’s awesome Phd candidate Kingsley Nweye for pulling off this monumental work🎓👏. And big thank you to EPRI & Siva Sankaranarayananfor providing the data and guidance. Connaisseurs of the subject matter will recognize that this research was based on the 2022 CityLearn Challenge dataset (stay tuned for the 2023 Challenge! #energystorage #energy ).

Our work unifies the areas of occupant-centric control (OCC) and grid-interactive efficient buildings (GEB), laying the foundation for further studies that explore the intersection between the two areas. We believe that our work will pave the way for more efficient, secure, and adaptable energy management systems. 🌍💡

Check out our paper for a deeper dive into our research and findings. We welcome your thoughts, questions, and discussions on this vital topic. 💪🌟

Repository to reproduce the experiments:

Zoltan Nagy
Zoltan Nagy
Assistant Professor

My research interests include reinforcement learning for buildings and smart cities.