A data-driven study of thermostat overrides during demand response events

Lucile Sarran*, H. Burak Gunay, William O'Brien, Christian Anker Hviid, Carsten Rode

*Corresponding author for this work

    Research output: Contribution to journalJournal articleResearchpeer-review

    Abstract

    In the context of increasing renewable energy penetration in energy systems, demand response is acknowledged as a solution to guarantee grid stability and security of supply. Direct control of appliances by utilities, however, may lead to user dissatisfaction and disengagement via overrides. The present study, based on data from 6,389 connected thermostats in North America in the summer of 2019, investigates users’ thermostat overriding behavior during demand response events targeting their air conditioners. An average event in this dataset was triggered around 3 p.m. and lasted three hours. The overall override rate was 12.9%. Overrides critically affected power usage during an event, with the share of the expected power demand reduction missed due to overrides being of the same order of magnitude as the override rate. In a decision tree analysis, the override rate showed to be particularly affected by occupants’ habitual setpoint change frequency, outdoor temperature, event duration, and occupants’ previous experience with demand response. Even though the dataset is not representative of all types of demand response events, this study highlights the potential lying in connected thermostat data for utilities to design tailored demand response events with an increased success rate and a smaller impact on occupant comfort.
    Original languageEnglish
    Article number112290
    JournalEnergy Policy
    Volume153
    Number of pages14
    ISSN0301-4215
    DOIs
    Publication statusPublished - 2021

    Keywords

    • Demand response
    • Connected thermostat
    • Occupant behavior
    • Override
    • Decision tree

    Fingerprint

    Dive into the research topics of 'A data-driven study of thermostat overrides during demand response events'. Together they form a unique fingerprint.

    Cite this