Predictive presence scheduling for a thermostat using machine learning

Patent metadata

Field	Value
Publication number	US-2022065478-A1
Application number	US-202117454451-A
Country	US
Kind code	A1
Filing date	Nov 10, 2021
Priority date	Dec 31, 2019
Publication date	Mar 3, 2022
Grant date	—

Abstract

A heating, ventilation, and air conditioning (HVAC) control device configured to generate the machine learning model using the first set of weights and the second set of weights. The machine learning model is configured to output a probability that a user is present at the space based on an input that identifies a day of the week and a time of a day. The device is further configured to determine a probability that a user is present at the space for a predicted occupancy schedule using the machine learning model, to determine an occupancy status based on a determined probability that a user is present at the space, and to set a predicted occupancy status in the predicted occupancy schedule based on a determined occupancy status for each time entry. The device is further configured to output the predicted occupancy schedule.

First claim

1 . A heating, ventilation, and air conditioning (HVAC) control device, comprising:
a memory operable to store:
an occupancy history log that comprises:
a plurality of occupancy statuses over a predetermined time period, wherein each occupancy status indicates one of a present status and an away status for a user within a space; and
a plurality of timestamps, wherein each timestamp is associated with an occupancy status from among the plurality of occupancy statuses; and

a predicted occupancy schedule comprising a plurality of time entries, wherein:
each time entry corresponds with a day of a week and a time of a day; and
each time entry is associated with a predicted occupancy status; and

a processor operably coupled to the memory, configured to:
generate a machine learning model using a first set of weights and a second set of weights, wherein:
the machine learning model is configured to output a probability that the user is present at the space based on an input that identifies a day of the week and a time of a day; and
generating the machine learning model comprises:
associating the first set of weights with days of the week based on the occupancy history log; and
associating the second set of weights with times of a day based on the occupancy history log;

determine a probability that the user is present at the space for each time entry in the predicted occupancy schedule using the machine learning model, wherein determining the probability that the user is present at the space comprises:
identifying a day and a time that is associated with a first time entry from the predicted occupancy schedule;
identifying a first weight value that corresponds with the identified day;
identifying a second weight value that corresponds with the identified time; and
computing the probability based on the first weight value and the second weight value;

determine an occupancy status for each time entry based on a determined probability that the user is present at the space corresponding with each time entry, wherein determining the occupancy status comprises:
selecting a second time entry;
comparing the determined probability that is associated with the second time entry to a threshold value; and
determining the occupancy status is a present status when the determined probability is greater than or equal to the threshold value;

set a predicted occupancy status for each time entry based on a determined occupancy status for each time entry; and
output the predicted occupancy schedule.

2 . The device of claim 1 , wherein the processor is further configured to:
determine a number of occupancy statuses in the occupancy history log; compare the number of occupancy statuses in the occupancy history log to a predetermined threshold value, wherein the predetermined threshold value corresponds with a minimum number of occupancy statuses for generating a machine learning model; and determine that the number of occupancy statuses in the occupancy history log is greater than the predetermined threshold value before determining the first set of weights for the machine learning model.

3 . The device of claim 1 , wherein:
the occupancy history log associates occupancy statuses with minutes of a day; and the processor is further configured to convert the occupancy history log to associate occupancy statutes with hours of a day before determining the first set of weights for the machine learning model.

4 . The device of claim 1 , wherein:
the processor is further configured to:
obtain historical weather information; and
determine a third set of weights for the machine learning model based on the historical weather information, wherein each of the third plurality of weights is associated with a temperature at a time of a day; and
the input for the machine learning model further identifies a forecasted temperature corresponding with a day of the week and a time of a day.

5 . The device of claim 1 , wherein outputting the predicted occupancy schedule comprises presenting the predicted occupancy schedule on a graphical user interface.

6 . The device of claim 1 , wherein outputting the predicted occupancy schedule comprises sending the predicted occupancy schedule to a thermostat for an HVAC system.

7 . The device of claim 1 , wherein outputting the predicted occupancy schedule comprises storing the predicted occupancy schedule in a memory.

8 . A predictive presence scheduling method, comprising:
generating a machine learning model using a first set of weights and a second set of weights, wherein:
the machine learning model is configured to output a probability that the user is present at the space based on an input that identifies a day of the week and a time of a day; and
generating the machine learning model comprises:
associating the first set of weights with days of the week based on an occupancy history log, wherein the occupancy history log comprises:
a plurality of occupancy statuses over a predetermined time period, wherein each occupancy status indicates one of a present status and an away status for a user within a space; and
a plurality of timestamps, wherein each timestamp is associated with an occupancy status from among the plurality of occupancy statuses; and

associating the second set of weights with times of a day based on the occupancy history log;

determining a probability that the user is present at the space for each time entry in a predicted occupancy schedule using the machine learning model, wherein:
the predicted occupancy schedule comprises a plurality of time entries, wherein:
each time entry corresponds with a day of a week and a time of a day; and
each time entry is associated with a predicted occupancy status; and

determining the probability that the user is present at the space comprises:
identifying a day and a time that is associated with a first time entry from the predicted occupancy schedule;
identifying a first weight value that corresponds with the identified day;
identifying a second weight value that corresponds with the identified time; and
computing the probability based on the first weight value and the second weight value;

determining an occupancy status for each time entry based on a determined probability that the user is present at the space corresponding with each time entry, wherein determining the occupancy status comprises:
selecting a second time entry;
comparing the determined probability that is associated with the second time entry to a threshold value; and
determining the occupancy status is a present status when the determined probability is greater than or equal to the threshold value;
setting a predicted occupancy status for each time entry based on a determined occupancy status for each time entry; and outputting the predicted occupancy schedule.

9 . The method of claim 8 , further comprising:
determining a number of occupancy statuses in the occupancy history log; comparing the number of occupancy statuses in the occupancy history log to a predetermined threshold value, wherein the predetermined threshold value corresponds with a minimum number of occupancy statuses for generating a machine learning model; and determining that the number of occupancy statuses in the occupancy history log is greater than the predetermined threshold value before determining the first set of weights for the machine learning model.

10 . The method of claim 8 , further comprising converting the occupancy history log to associate occupancy statutes with hours of a day before determining the first set of weights for the machine learning model.

11 . The method of claim 8 , further compris

Assignees

• Lennox Ind Inc

Inventors

• Venkatesh Sridhar
• Delgoshaei Payam
• Manohararaj Janathkumar

Classifications

• F24F11/64

  using pre-stored data · CPC title

• F24F2130/10

  Weather information or forecasts · CPC title

• F24F2110/10

  Temperature · CPC title

• F24F2120/12

  Position of occupants · CPC title

• F24F11/65

  for selecting an operating mode · CPC title