Detecting appliances in a building from coarse grained meter data with partial label

A method for detecting appliance in a building, comprising:

• receiving meter data associated with energy consumption, the meter data comprising at least energy consumption data associated with usage of the appliance;
• receiving customer data associated with the meter data;
• extracting features, by a processor, for training a model for detecting the appliance, based on at least the meter data and the customer data;
• computing empirical features and statistical features from the extracted features, wherein the empirical features include temperature dependent heating consumption, ratio between average consumption during heating required days and during non-heating required days, and the statistical features comprising at least one of discrete Fourier transform coefficients or discrete wavelet transform coefficients;
• constructing, by the processor, the model at least based on the empirical and statistical features;
• and
• applying the model to candidate meter data associated with a candidate customer, for detecting whether the candidate customer is using the appliance,
• wherein the customer data comprises data associated with customers known to use the appliance and data associated with customers unknown to use or not use the appliance, and wherein the model is built based on the features extracted from the data associated with customers known to use the appliance and the data associated with customers unknown to use or not use the appliance.