Register today for SyracuseCoE's November Research & Technology Forum!

Presentation: Intelligent Decision-Making for Building Decarbonization

Speaker: Dr. Bianca Howard, Assistant Professor of Mechanical Engineering, Columbia University

In this talk, Dr. Howard will share research from the Building Energy Research Laboratory (BERL) that brings together building physics, optimization, and artificial intelligence to improve how buildings are designed and operated for decarbonization. BERL work ranges from large-scale policy and portfolio planning to real-time control of heating and cooling systems, with a focus on developing approaches that are not only technically effective but also address current real-world challenges.

At the urban scale, Howard’s team uses multi-objective optimization to develop decarbonization pathways that consider not only cost and greenhouse gas emissions but also broader social goals. In a study of buildings in West Harlem, New York City, her team compared retrofit strategies that minimize cost and emissions with those that also include metrics such as energy burden and job creation. The results show that incorporating these additional objectives leads to very different retrofit portfolios, emphasizing the need for decision-support tools that make these trade-offs visible and explicit.

At the building scale, the lab investigates how accurately simulators need to represent physical systems to train reinforcement learning controllers for HVAC operation. Reinforcement learning shows promise as an intelligent control method due to the reduced need for online computation; however, these models must be trained in simulations or with data in advance. Focusing on simulations, the team explored training these agents using several EnergyPlus models with different thermal properties. Their work indicates that imprecise models can be used to train agents that result in good control performance. This potentially indicates that only site visits would be needed to develop intelligent control agents, meaning agents could be generated in one day instead of relying on weeks of historical data.

Together, these studies show how building physics, optimization, and AI can come together to create practical, scalable solutions for building decarbonization across design, policy, and operational contexts.