Evaluating Thermostat Demand Response Under Limited Telemetry: Window Sensitivity and Conclusion Stability
Demand-response (DR) evaluation depends on the observation window, but thermostat labels may not identify complete dispatch boundaries. We examined how active-label and fixed-duration windows affect cooling-runtime response estimates, uncertainty-qualified categories, and descriptive household comparisons in cleaned 2017 ecobee data. Three past-only baselines were evaluated using a forward calendar split. Among 175 common September onsets from 59 homes, the matched mean response was 62.72 s per five-minute step lower in the four-hour than in the active-label window (approximate household-by-date 95% interval for the difference: [−99.57,−29.29]). Across models, 30.3–45.1% of onsets changed category, predominantly becoming unresolved rather than reversing a resolved sign. Upper-quarter household membership also changed across windows, including in a balanced repeat-home analysis. Non-DR references and split-half checks qualify these exploratory comparisons. Uncertain dispatch ends and zero-fill ambiguity preclude causal interpretation. The findings support explicit reporting of window definitions and conclusion stability, without validating household targeting, override attribution, or measured energy savings.
Authors
- Mingzhi Yang (ORCID: https://orcid.org/0000-0002-8840-2164)
- Lei Zhang
- Yue Liu
Institutions
- China University of Mining and Technology (CN)
Publication Details
- Journal
- Energies
- Published
- 2026-11-16
- DOI
- https://doi.org/10.3390/en19225524
- Primary Topic
- Smart Grid Energy Management
- Type
- article
- Field-Weighted Citation Impact
- 0.00