UC Berkeley

Thermostats play a critical role in residential demand flexibility; their performance depends not only on algorithms but on how occupants schedule, override, and adapt their systems. Current grid-interactive models employ idealized control patterns, overlooking the behavioral variability that drives actual HVAC operation. We studied thermostat scheduling behavior using nine years (2017–2025) of Ecobee Donate-Your-Data telemetry from 1,263 California households. Across the study period, households used 517 unique schedule names, including the three Ecobee defaults, Home, Sleep, and Away, and 514 user-defined labels. We classified these names into a rule-based taxonomy with 8 main categories and 19 subcategories, then grouped households by schedule configuration complexity, naming patterns, and setpoint calibration. These patterns yielded five behavioral archetypes, four statistically robust: the Passive Default (63.4%, factory defaults with minimal thermal differentiation), the Active Customizer (29.5%, progressive grid-interactive schedule expansion), the Disciplined Minimalist (3.2%, deliberate wide thermal deadband), the Cold-Climate Customizer (2.2%, extreme nocturnal setback), and the Seasonal Switcher (0.4%, two-program seasonal alternation).

Schedule semantic content independently predicts heating and cooling setpoints and HVAC runtime after controlling for archetype, year, and outdoor temperature, establishing that schedule naming encodes behavioral intention. These archetypes provide an empirically grounded basis for behavior-aware Demand Response program design and establish schedule name analysis as a scalable behavioral segmentation tool. The vocabulary households assign to their schedules is itself a behavioral record, one that predicts HVAC outcomes independently of archetype and weather and is already embedded in existing utility datasets without surveys or additional instrumentation.