Disaggregation of appliance usage from electrical meter data

A method for determining presence of an appliance within a surveyed household in which there are no sensors for measuring consumption of specific appliance, said method implemented by one or more processing devices operatively coupled to a non-transitory storage device, on which are stored modules of instruction code that when executed cause the one or more processing devices to perform:

• pre-processing of historical consumption of appliances in sampled households based on actual measurements by sensors associated with said appliances in relation to characteristics of the households including one or more of lifestyle of the occupants and environmental time-dependent parameters;
• applying a learning algorithm for identifying first statistical correlations between presence of appliances in each sampled household and at least one of 1) household profile parameters, 2) household actual periodic consumption patterns, and 3) household actual periodic consumption usage pattern in relation to environmental time-dependent parameters;
• building household appliance presence model which defines the relation between the presences of each appliance and the household profile consumption usage based on identifying first statistical correlations;
• and
• determining the probability of each appliance presence at the surveyed household using identified profile parameters and the calculated consumption of the actual behavior pattern of the surveyed household, based on processing identified first statistical correlations between presence of appliances at each household and at least one of:
• 1) household profiles parameters, 2) household actual periodic consumption usage pattern, 3) household actual periodic consumption usage pattern in relation to environmental time-dependent parameters, wherein the surveyed household is not part of the sampled households;
• wherein the data in the sampled households include actual activation and consumption across the day of each appliance, overall data of the household, household profile and characteristics, historical consumption reads for the appliances and household overall, weather condition across the day;
• and further comprising:
• determining, from the first statistical correlations and identified presence of appliances generated by the statistical appliance presence model, a machine learning appliance activation model;
• and
• determining, from parameters of the surveyed household and the machine learning appliance activation model, a probability of an appliance's activation in the surveyed household;
• re-estimating a detected daily activated list of all appliances in the household using the machine learning appliance activation model based on consumption usage per period;
• applying the machine learning algorithm for identifying second statistical correlations between household actual periodic portion, hourly or sub-hourly consumption usage pattern and actual activation per periodic portion of each appliance in relation to at least one of 1) environmental time dependent parameters or life style of the occupant, 2) household profile and actual consumption rate of the household overall, and 3) data on historical usage in former days or hours of the household overall;
• building a household appliance hourly and/or sub-hourly activation model based on identified machine learning second statistical correlations, which defines the statistical correlations between the hourly or sub-hourly activation of each appliance and the household profile consumption usage, and detected daily activation of appliances generated by the machine learning appliance activation model;
• and
• applying the household appliance hourly and/or sub-hourly machine learning activation model for determining for each identified appliance at what hours the detected appliance was activated during detected activated period at the surveyed household by processing identified machine learning second statistical correlations between appliance activation and at least one of profile parameters, actual behavior pattern of the analyzed household in relation to actual time dependent environmental parameters of the relevant time period, and consumption pattern and daily activation in former days or hours;
• and
• wherein the household appliance presence model, the machine learning appliance activation model, and the household appliance hourly and/or sub-hourly machine learning activation model are configured, trained, and applied to different time periods, different tested features, and different analyzed parameters.

A computer based system for determining presence appliance within a household in which there are no sensors for measuring consumption of specific appliance, said system comprising a non-transitory storage device and one or more processing devices operatively coupled to the storage device on which are stored modules of instruction code executable by the one or more processors:

• a history analysis module configured to pre-processing per meter of sample households of historical consumption of appliances based actual measurement performed by sensors associated with said appliances in relation to profile of household including characteristics of the household and/or lifestyle of the occupant and environmental time dependent parameters;
• an appliance detection module for
• applying learning algorithm for identifying first statistical correlations between presence of appliances at each household and at least one of 1) household profiles parameters, 2) household actual periodic consumption pattern, and 3) household actual periodic consumption pattern in relation to environmental time dependent parameters;
• building household appliance presence model which defines the relation between the presences of each appliance and the household profile consumption based on identifying first statistical correlations;
• and
• determining the probability of each appliance presence at the surveyed household using identified profile parameters and the calculated consumption of the actual behavior pattern of the surveyed household, based on processing identified first statistical correlations between presence of appliances at each household and household profiles parameters, household actual periodic consumption pattern in relation to environmental time dependent parameters, wherein the surveyed household is not part of the sampled households;
• wherein the data in the sampled households include actual activation and consumption across the day of each appliance, overall data of the household, household profile and characteristics, historical consumption reads for the appliances and household overall, weather condition across the day;
• and further comprising:
• determining, from the first statistical correlations and identified presence of appliances generated by the statistical appliance presence model, a machine learning appliance activation model;
• and
• determining, from parameters of the surveyed household and the machine learning appliance activation model, a probability of an appliance's activation in the surveyed household;
• re-estimating a detected daily activated list of all appliances in the household using the machine learning appliance activation model based on consumption usage per period;
• applying the machine learning algorithm for identifying second statistical correlations between household actual periodic portion, hourly or sub-hourly consumption usage pattern and actual activation per periodic portion of each appliance in relation to at least one of 1) environmental time dependent parameters or life style of the occupant, 2) household profile and actual consumption rate of the household overall, and 3) data on historical usage in former days or hours of the household overall;
• building a household appliance hourly and/or sub-hourly activation model based on identified machine learning second statistical correlations, which defines the statistical correlations between the hourly or sub-hourly activation of each appliance and the household profile consumption usage, and detected daily activation of appliances generated by the machine learning appliance activation model;
• and
• applying the household appliance hourly and/or sub-hourly machine learning activation model for determining for each identified appliance at what hours the detected appliance was activated during detected activated period at the surveyed household by processing identified machine learning second statistical correlations between appliance activation and at least one of profile parameters, actual behavior pattern of the analyzed household in relation to actual time dependent environmental parameters of the relevant time period, and consumption pattern and daily activation in former days or hours;
• wherein the household appliance presence model, the machine learning appliance activation model, and the household appliance hourly and/or sub-hourly machine learning activation model are configured, trained, and applied to different time periods, different tested features, and different analyzed parameters.