(CS3.3) Data – driven approaches to deep retrofits and their impact on heritage buildings

For the building sector to meet its emissions reduction targets, it is estimated that 3% of the existing building stock needs to be retrofitted each year. Presently in Canada, the annual retrofit rate is hovering around 1%/year, well short of the target. There are numerous barriers which are limiting the retrofit rate. The technical barriers of deep retrofits have largely been overcome, and it is the non-technical barriers such as financial, socioeconomic, political, and informational barriers which remain.
In this presentation, I will discuss how we are developing data-driven models to help address the low retrofit adoption rate considering both technical and non-technical factors, and what implications this may have on heritage buildings. The data-driven models being developed utilize machine learning techniques such as regression, neural networks, surrogate modelling and are underpinned by data collected from a number of sources depending on building archetype. The machine learned models are being trained for a number of relevant applications. First is to predict optimal energy efficiency measure bundles based on key inputs such as geometry, envelope parameters, HVAC system, past performance etc. Second is it to streamline the process of executing a deep retrofit for building owners, simplifying schematic design, and providing more accurate cost estimates. Third is to make financially viable business cases for deep retrofits, especially for lower-income owners, whose buildings often offer the greatest scope for improvements. Part of this equation includes how to capture the difficult to monetize benefits of deep retrofits, such as improved occupant health and well-being, improved climate resilience, or maintaining heritage sites for future generations.
It is important to emphasize that these models are underpinned by data, and that data tends to be dominated by contemporary construction. Many heritage buildings are considered as edge cases because they have rarer assemblies, systems, and architectural features. Therefore they are poorly represented in the data and this will not allow them to reap the benefits of data-driven approaches, and may expose them to inappropriate solutions and advice which can have negative consequences if followed blindly. For heritage buildings to be properly represented in these models we must have sufficient data from them. Data-driven and AI approaches are entering the building sector, and it is important that the heritage community understand and embrace it with sharing performance and technical data. This includes undertaking measurement & verification procedures to report the energy consumption before and after a retrofit to help us understand what interventions work best, and to openly share building data, retrofit measures, costs, and performance data using a complete and standardized template. This will enable us to compare and contrast retrofit approaches to better train our models.