When a building is retrofitted, the energy savings promised and those actually observed do not coincide — the performance gap. This three-year postdoctoral fellowship, run with a Québec industrial partner in energy retrofit, takes on that gap: building a knowledge graph that unifies data from real projects, grafting interpretable machine learning models onto it, and producing forecasts corrected for optimism bias by grounding them in comparable past projects rather than in simulation alone.
Profile sought
A doctorate in computer, electrical or building engineering, or in applied data science. Semantic web — RDF, SPARQL, SHACL, ontology engineering — and comfortable with Python and a graph database. In machine learning, interpretability is a requirement here rather than a refinement: a forecast has to be explainable to a client. Familiarity with building ontologies (Brick, Haystack, ASHRAE 223P), building management data (BACnet, Modbus), BIM models or measurement and verification transfers directly. French or English is required; the work is published in English.
Funding
Mitacs Accelerate funding over thirty-six months, in units of four to six months, with an industrial partner. The work is shared between the laboratory and the company, which opens access to the history of its projects and to its thermal and measurement-and-verification specialists.
Supervision
Joint supervision: scientific guidance at LIREI (UQTR), technical mentoring at the industrial partner. Fortnightly meetings and a presentation of the work at the end of each internship unit.