Low frequency energy disaggregation techniques

1. A method of performing appliance itemization based on consumption data for a whole house, comprising:

receiving at a processor the consumption data;

determining by the processor if the consumption data includes active signals and/or inactive signals;

upon detection of an active signal:

detecting by the processor, based at least in part on the active signal, active water heating, and if active water heating is detected, estimating active water heating consumption;

detecting by the processor, based at least in part on the active signal, lighting, and if lighting is detected, estimating lighting consumption;

upon detection of an inactive signal:

detecting by the processor, based at least in part on the inactive signal, passive water heating, and if detected, estimating passive water heating consumption;

detecting by the processor, based at least in part on the inactive signal, refrigeration, and if refrigeration is detected, estimating refrigerator consumption;

detecting by the processor, based at least in part on the inactive signal, vacation mode.

2. The method of claim 1 , wherein the consumption data is received from a utility.

3. The method of claim 1 , wherein the consumption data is received from an advanced metering infrastructure (AMI) device.

4. The method of claim 1 , wherein the consumption data is of a low resolution comprising data received at fifteen (15) minute, thirty (30) minute, or sixty (60) minute intervals.

5. The method of claim 1 , wherein the processor determines if the consumption data includes active signals and/or inactive signals based at least in part upon:

using a sliding window function and storing the energy consumption in a vector;

subsequently comparing the vector with lowest energy consumption windows to determine inactivity.

6. The method of claim 5 , wherein the processor estimates active water heater consumption by estimating a typical amplitude of a water heater, behavior modeling, and data smoothing.

7. The method of claim 6 , wherein behavior modeling comprises user-specific patterns are modeled, including attributes of time of day and day of week.

8. The method of claim 7 , wherein attributes may further include characteristics selected from the group consisting of temperature, season, number of occupants of the house, specific appliance usage, and demographic data including geographic region, age of occupants, gender of occupants, and/or socioeconomic status of occupants.

9. The method of claim 1 , wherein the processor detects lighting based at least in part by identifying arcs of consumption in time periods where lighting is expected to be used, wherein identifying arcs of consumption comprises:

removing unsuitable data points from the consumption data;

cleaning the consumption data using percentile filtering, minimum value removal, and/or smoothing noise removal;

detecting and removing seasonal loads by selecting time points showing seasonal characteristics including contiguous days of absence or presence when typically occurring lighting is absent;

extracting a lighting band, comprising a collection of times in a day at which lighting is present.

10. The method of claim 9 , wherein the processor estimates lighting consumption at each data point along with lighting capacity parameters, timings of lighting, and seasonality.

11. The method of claim 1 , wherein the processor detects passive water heating by amplitude estimation, frequency estimation, and temperature correlation.

12. The method of claim 11 , wherein the processor estimates passive water heating consumption by estimating typical amplitude and frequency of passive pulses, determined by a power rating of a detected water heater and the typical duration of passive pulses, wherein the duration of passive pulses is a function of a temperature setting of the water heater, insulation of the water heater, and the outside temperature.

13. The method of claim 1 , wherein the processor detects refrigeration by analyzing data during low activity periods in which regular pulses of a refrigeration compressor motor are detectable.

14. The method of claim 13 , wherein the processor estimates refrigerator consumption by:

analyzing several months of consumption data and identifying low activity periods;

scoring each low activity period for quality;

compiling low activity periods that meet a quality threshold;

using a machine learning model to, based on the low activity periods that meet the quality threshold, predict refrigeration consumption.

15. The method of claim 14 , wherein the machine learning model is a regression model based at least in part on training data.

16. The method of claim 1 , wherein the processor detects vacation mode by identifying multiple consecutive days of static, lower-than-average consumption.

17. The method of claim 16 , wherein identifying multiple consecutive days of static, lower-than-average consumption comprises:

reviewing consumption data spanning the previous three (3) weeks;

calculating a baseload based on previous consumption data and setting a threshold;

calculating day energy consumption, and determining if day energy consumption is within the threshold;

if the day energy is within the threshold, calculate energy consumption based at least in part on the use of a sliding window function, with a window size of approximately four (4) hours by two (2) hours, and storing the energy consumption in a vector;

determine if the size of the vector is greater than one day, and if so, determine the standard deviation for the vector;

determine if the standard deviation for the vector is within the threshold, and if so, identify the top three deviated values from the mean of the vector;

determine if the top three deviated values from the mean of the vector are within the threshold, and if so determine the mean of the vector;

determine if the mean of the vector is within the threshold;

if the mean of the vector is within the threshold, a vacation is detected with start and end time as the start and end time of the vector.

18. The method of claim 1 , further comprising:

providing to a customer an itemization of detected appliances and estimated consumption for each detected appliance.

19. A method of performing appliance itemization based on consumption data for a whole house received from an advanced metering infrastructure (AMI) device, the consumption data being of a low resolution comprising data received at fifteen (15) minute, thirty (30) minute, or sixty (60) minute intervals, the method comprising:

receiving at a processor the consumption data;

determining by the processor if the consumption data includes active signals and/or inactive signals;

upon detection of an active signal:

detecting by the processor, based at least in part on the active signal, active water heating, and if active water heating is detected, estimating active water heating consumption at least in part by estimating a typical amplitude of a water heater, behavior modeling, and data smoothing, wherein behavior modeling comprises user-specific patterns are modeled, including attributes of time of day and day of week;

detecting by the processor, based at least in part on the active signal, lighting based at least in part by identifying arcs of consumption in time periods where lighting is expected to be used, wherein identifying arcs of consumption comprises:

removing unsuitable data points from the consumption data;

cleaning the consumption data using percentile filtering, minimum value removal, and/or smoothing noise removal;

detecting and removing seasonal loads by selecting time points showing seasonal characteristics including contiguous days of absence or presence when typically occurring lighting is absent;

extracting a lighting band, comprising a collection of times in a day at which lighting is present;

if lighting is detected, estimating lighting consumption at each data point along with lighting capacity parameters, timings of lighting, and seasonality;

upon detection of an inactive signal:

detecting by the processor, based at least in part on the inactive signal, passive water heating based at least in part on amplitude estimation, frequency estimation, and temperature correlation, and if detected, estimating passive water heating consumption at least in part by estimating typical amplitude and frequency of passive pulses, determined by a power rating of a detected water heater and the typical duration of passive pulses, wherein the duration of passive pulses is a function of a temperature setting of the water heater, insulation of the water heater, and the outside temperature;

detecting by the processor, based at least in part on the inactive signal, refrigeration at least in part by analyzing data during low activity periods in which regular pulses of a refrigeration compressor motor are detectable, and if refrigeration is detected, estimating refrigerator consumption at least in part by:

analyzing several months of consumption data and identifying low activity periods;

scoring each low activity period for quality;

compiling low activity periods that meet a quality threshold;

using a regression machine learning model to, based on the low activity periods that meet the quality threshold, predict refrigeration consumption; and

detecting by the processor, based at least in part on the inactive signal, vacation mode by identifying multiple consecutive days of static, lower-than-average consumption; and

providing to a customer an itemization of detected appliances and estimated consumption for each detected appliance.

20. A method of performing appliance itemization based at least in part on consumption data for a whole house received from an advanced metering infrastructure (AMI) device, the consumption data being of a low resolution comprising data received at fifteen (15) minute, thirty (30) minute, or sixty (60) minute intervals, the method comprising:

applying disaggregation models to the consumption data to provide detection and estimation of any present lighting appliances, water heating appliances, refrigeration appliances, pool pumps, heating appliances, or cooling appliances;

applying rule-based models to the consumption data to provide detection and estimation of any present cooking appliances, laundry appliances, entertainment devices or appliances, and miscellaneous appliances;

wherein the disaggregation models and the rule-based models provide for a complete or near complete appliance level itemization and estimation for the whole house.