Self-learning, real-time, data-driven power metering system

• US 10,396,555 B2
• Filed: 05/26/2017
• Issued: 08/27/2019
• Est. Priority Date: 07/14/2015
• Status: Active Grant

• Current Assignee: REstore NV (Centrica PLC)
• Original Assignee: REstore NV (Centrica PLC)
• Inventors: Rombouts, Jan-Willem
• Primary Examiner(s): Pan, Yuhui R
• Application Number:
  US15/606,526
• Publication Number:
  US 20170264095A1
• Time in Patent Office: 823 Days
• CPC Class Codes: G05B 17/02 electric
• CPC Class Codes: G06Q 10/04 Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem" market predictions or forecasting for commercial activities G06Q30/0202
• CPC Class Codes: G06Q 50/06 Energy or water supply
• CPC Class Codes: H02J 3/00 Circuit arrangements for ac mains or ac distribution networks
• CPC Class Codes: Y02P 80/10 Efficient use of energy, e.g. using compressed air or pressurized fluid as energy carrier

Abstract

A power metering system includes any number of sites and a central database. Loads within a site are unmetered and a sensor data vector over time includes sensor data, operational data and external data. Iteration occurs over time intervals for all unmetered loads and the unmeasured power for each interval is disaggregated by matching a load type with a computer model in the central database having the same feature domain in order to predict the power usage of each load. Iteration again occurs over those intervals in which only a single load is operating; the power of that load is determined to be the unmeasured power minus miscellaneous power, and a new computer model is fitted for that single load and also uploaded to the database. The feature domain of an existing model may also be increased. Both iteration steps repeat until no more unmeasured power can be disaggregated.

Claims

1. A method of addressing disaggregation error at a site, said method comprising:

computing a disaggregation error of a plurality of loads over a plurality of time intervals at said site, said computing using measured power of said loads and predicted power of said loads, said measured power being measured using temporary power meters;

determining that said disaggregation error is not within a requirement of said site;

ranking said loads based upon a length of time during which each of said loads is operating and no power model is available for said each of said loads;

adding a permanent power meter to a worst load of said ranking having a longest time of said ranking; and

measuring a power of said worst load using said added permanent power meter, whereby said disaggregation error is reduced.

2. A method as recited in claim 1 further comprising:

performing said step of adding a permanent power meter iteratively load-by-load until said disaggregation error is within said requirements of said site.

3. A method as recited in claim 1 further comprising:

iterating over said plurality of loads and disaggregating power of said site in order to predict power of said loads over said time intervals, said iterating occurring before said computing.

4. A method as recited in claim 1 wherein said computing uses a cost based upon said measured power of said loads and a cost based upon said predicted power of said loads.

5. A method of addressing disaggregation error at a site, said method comprising:

computing a disaggregation error of a plurality of loads over a plurality of time intervals at said site, said computing using measured power of said loads and predicted power of said loads, said measured power being measured using temporary power meters;

determining that said disaggregation error is not within a requirement of said site;

ranking said loads based upon a length of time during which each of said loads is operating and no power model is available for said each of said loads;

adding a data sensor to a worst load of said ranking having a longest time of said ranking; and

predicting a power of said worst load using a power model that uses input from said data sensor, whereby said disaggregation error is reduced.

6. A method as recited in claim 5 further comprising:

performing said step of adding a data sensor iteratively load-by-load until said disaggregation error is within said requirement of said site.

7. A method as recited in claim 5 further comprising:

iterating over said plurality of loads and disaggregating power of said site in order to predict power of said loads over said time intervals, said iterating occurring before said computing.

8. A method as recited in claim 5 wherein said computing uses a cost based upon said measured power of said loads and a cost based upon said predicted power of said loads.

9. A method as recited in claim 5 further comprising:

predicting said power by increasing a feature domain of said power model to utilize data input from said data sensor.

10. An apparatus for addressing disaggregation error at a site, said apparatus comprising:

a plurality of loads operating at said site;

a computer arranged to

compute a disaggregation error of said loads over a plurality of time intervals using measured power of said loads and predicted power of said loads, said measured power being measured using temporary power meters,

determine that said disaggregation error is not within a requirement of said site, and

rank said loads based upon a length of time during which each of said loads is operating and no power model is available for said each of said loads; and

a permanent power meter added to a worst load of said ranking having a longest time of said ranking, said permanent power meter arranged to measure a power of said worst load, whereby said disaggregation error is reduced.

11. An apparatus for addressing disaggregation error at a site, said apparatus comprising:

a plurality of loads operating at said site;

a computer arranged to

compute a disaggregation error of said loads over a plurality of time intervals using measured power of said loads and predicted power of said loads, said measured power being measured using temporary power meters,

determine that said disaggregation error is not within a requirement of said site, and

rank said loads based upon a length of time during which each of said loads is operating and no power model is available for said each of said loads; and

a data sensor added to a worst load of said ranking having a longest time of said ranking, said data sensor arranged to input data to a power model that predicts power of said worst load, whereby said disaggregation error is reduced.