Occupancy is the primary driver of commercial building energy consumption, yet systematic, sensor-free occupancy inference at portfolio scale remains an open challenge. Physical occupancy sensing infrastructure requires substantial capital investment, limiting deployment to premium assets. Smart meters, in contrast, are universally deployed across the commercial sector and provide granular half-hourly consumption data that implicitly encodes building activity levels. This paper presents a Gaussian Mixture Model (GMM) framework for inferring occupancy states from half-hourly electricity consumption patterns, applied to 1,051 commercial buildings from the Building Data Genome Project 2. The GMM identifies two latent states per building, specifically occupied and unoccupied, by fitting a bimodal mixture to the half-hourly consumption distribution. Buildings with near-unimodal distributions (61 buildings, 5.5%) are identified and excluded from waste quantification with explicit discussion of their characteristics. Applied across the remaining 990 buildings, the framework estimates a mean occupancy rate of 46.5% of half-hourly intervals, with substantial variation across building types (education 38.2%; healthcare 71.4%; office 52.1%). Non-occupancy energy consumption accounts for an estimated annual waste of £2.436 billion across the portfolio, equivalent to 23.1% of total consumption, extrapolating to an estimated £40–120 billion across the UK commercial estate (upper-bound: £128.2 billion). GMM estimates are validated against known occupancy schedules for education (89.7% holiday-period non-occupancy match) and office buildings (94.3% weekend non-occupancy match). These results demonstrate that smart meter data, analysed with appropriate probabilistic models, provides actionable occupancy intelligence at portfolio scale without physical sensor infrastructure.
Olajide Ayoola (2026) studied this question.