Canadian PI: Dr. Elie Azar
Canadian Institution: Carleton University
Indian PI: Dr. Albert Thomas
Indian Institution: Indian Institute of Technology (IIT) Bombay
Project Summary:
The building sector is a major contributor to global energy consumption, accounting for roughly 55% of all energy use and 28% of energy-related CO2 emissions. For India and Canada to achieve their net-zero emission goals by 2070 and 2050 respectively, urgent action is needed to transition to a less energy-intensive building stock, which could be achieved through net-zero energy buildings (NZEBs). A NZEB is defined as structures built to generate enough energy, either on-site or off-site, to meet at least their own energy needs. To test and experiment with different NZEB designs, researchers commonly employ Building Energy Simulation (BES) to predict building performance based on various inputs (e.g., building design features, operation patterns, etc.).
BES tools can emulate building performance with high accuracy but are computationally intensive, limiting their applicability. To address this gap, researchers have turned to Machine Learning (ML) based surrogate models to complement BES and overcome computational limitations. Using data generated by a BES model (e.g., from 1000 runs), an ML model can be trained and tested to mimic BES capabilities (e.g., energy prediction) and subsequently test many building design scenarios (e.g., 100,000) at a very low computational cost. Previous studies demonstrated proofs-of-concept of the hybrid ML-BES approach. However, these studies present significant barriers to the readiness of the approach for complex design problems, particularly for NZEBs. The limitations include: (i) limited validation of BES models with real building data, (ii) limited adaptation of models to optimisation problems, especially in the context of NZEB, and (iii) limited application of methods to multiple buildings to ensure generalizability.
This project directly addresses these gaps by proposing a Simulation-based Machine Learning Optimisation (SML-Opt) framework to guide the design and operation of NZEBs, with a specific focus on retrofitting existing buildings. An integral part of the project is demonstrating and validating the framework on buildings in Canada and India, confirming its applicability to different contexts, building characteristics, and climate conditions. The case studies would lead to focused recommendations on how the building regulations of both countries could be adjusted to support the net-zero transition of their building sectors.
