US Patent 9800958 Training power models using network data

Power models may be used to identify devices in a building or state changes of devices in a building by processing a power monitoring signal with the power models. A power model for a device may be trained using examples of power monitoring signals corresponding to that device, and network monitoring may be used to identify training data for training power models for that device. Network monitoring may include receiving information about network packets transmitted by devices in the building and processing the information about the network packets to determine information about a state change of a first device in the building at a first time. A portion of the power monitoring signal that includes the first time may then be used to train a power model for the first device. The trained power model for the first device may then be deployed to a building where it used to perform power monitoring of the first device.