Apparatus and method for non-intrusive load monitoring aided hybrid high and low frequency approaches

1. An apparatus for monitoring power consumption of individual appliances in a building, comprising:

a smart meter circuit configured for measuring and collecting power level data across three electrical supply phases and compatible with three-phase services; and

a control circuit configured for

data handling,

sensing an activity level in said building,

switching between high frequency sampling and low frequency sampling modes based on a sensed activity level in said building, and

sending data to a cloud based server,

wherein said control circuit is configured to switch to high frequency sampling when the sensed activity level is above a specified threshold, and wherein high frequency data sampled during high frequency sampling has extractable voltage current (VI) trajectory data, and

wherein said control circuit is configured to switch to low frequency sampling when the sensed activity level is below said specified threshold, and wherein said power level data is measured and collected during said low frequency sampling,

wherein said activity level sensed is based on all of a number of appliance events in a window of time, a number of occupants in the building during a sensing period, and a time of day,

wherein the window of time, the sensing period, and the time of day, may be the same or different.

2. The apparatus of claim 1 wherein detection of the number of occupants is performed using a wireless network which communicates with mobile phones carried by individual occupants.

3. The apparatus of claim 1 wherein said control circuit is configured for

connecting to a WI-FI router,

controlling one or more devices connected to said WI-FI router, and

controlling a secure communication between a mobile application,

the cloud based server, and the smart meter circuit.

4. A method for monitoring power consumption of individual appliances in a building, comprising:

sensing an activity level in said building;

using a control circuit to switch between high frequency sampling and low frequency sampling modes based on the sensed activity level, wherein said high frequency sampling is performed when the activity level in said building is above a specified threshold and wherein said low frequency sample is performed when the activity level in said building is below said specified threshold;

wherein said activity level sensed is based on all of a number of appliance events in a window of time, a number of occupants in the building during a sensing period, and a time of day,

wherein the window of time, the sensing period, and the time of day may be the same or different;

sending high frequency data sampled during said high frequency sampling mode to a cloud based service for analysis, wherein said high frequency data has extractable voltage current (VI) trajectory data; and

using a smart meter circuit compatible with three-phase services to measure and collect power level data across three electrical supply phases during low frequency sampling.

5. A method of learning the energy consumption signatures of individual appliances in a building, comprising:

determining an activity level in the building by each of

measuring a number of appliance activation or deactivation events over a given time,

measuring a number of occupants of the building during the given time, and

using a machine learning algorithm to identify high activity times of day;

switching, based on a specified activity threshold, between

sampling power consumption at a high frequency during periods of high activity in the building, and

sampling power consumption at a low frequency during periods of low activity in said building;

receiving, during sampling power consumption at said high frequency, at a cloud based server high frequency sampling data;

processing at said cloud based server said high frequency sampling data with a convolutional neural network to disaggregate energy consumption signatures of individual appliances;

receiving, during sampling power consumption at said low frequency, at a client side smart meter the low frequency sampling data; and

processing at said client side smart meter said low frequency sampling data with a convolutional neural network to predict a midpoint energy consumption of individual appliances.

6. The method of claim 5 wherein said processing at said cloud base server includes extracting voltage current (VI) trajectory data from said high frequency sampling data.