A non-transitory machine-readable storage medium storing one or more sequences of instructions for load recognition from a monitored signal and application to energy management service, which instructions, when executed by one or more processors, cause the one or more processors to carry out the steps of:
- extracting, by a feature extraction algorithm processor, features from a set of power data using a feature extraction algorithm to obtain a plurality of events, wherein the set of power data reflects power consumed by one or more loads;
- initializing, by an initializing processor, a load library repository with initial load instance data, the initial load instance data comprising:
- initial estimates of state probabilities and state transition probabilities that represent a plurality of loads;
- performing a load recognition algorithm, by a load recognition algorithm processor, said load recognition algorithm comprising the steps of:
- for an event x from said plurality of events:
- assigning, by an assigning processor, said event x and particular subsequent events to a particular load,
- wherein the assigning is based on said particular load giving a best posterior probability from a best posterior probability algorithm performed on said plurality of events, and
- wherein said assigning is performed after performing a search and classification algorithm on said plurality of events;
- and
- adding, by an adding processor, load instance data associated with said particular load to said load library repository and continuing the adding load instance data for a next available event from said plurality of events;
- and
- updating based on said adding of load instance data, by an updating processor, said load library repository with new state probabilities and state transition probabilities,
- repeating said load recognition algorithm using the updated load library repository until a particular criterion is met;
- and
- using, by an energy management service processor, said added load instance data in providing energy consuming feedback and usage analysis to be used for energy management service;
- wherein said updated load library repository is used in a continuous mode on a continuous stream of power data from the plurality of loads and wherein when said added load instance data is added, combining said added load instance data with said initial load instance data to compute said new state probabilities and state transition probabilities from said combination;
- said continuous mode further comprising the steps of:
- extracting in real time, by said feature extraction algorithm, features from a window of said stream of power data for eligible events of said plurality of events in said window, performing a roughly estimated assignment to the particular load based on a state and state transition probability of each eligible event only;
- if there are not enough subsequent events for each roughly assigned event from said window and past windows of power data, then return to said extracting in real time step, otherwise:
- performing load recognition, by said load recognition algorithm, to a roughly assigned event with enough subsequent events based on a highest posterior probability search and classification and creating load instance recognition data;
- marking, by a marking processor, said roughly assigned event with said enough subsequent events as used;
- adding, by said adding processor, said load instance recognition data to said load library repository;
- updating based on the added load instance recognition, by said updating processor, state and state transition probability estimates for said particular load data;
- and
- with the arrival of a new window of power data, repeating said steps beginning from said step of extracting in real time features from said window of stream of power data;
- wherein said plurality of loads comprise a plurality of appliances.