A new transient feature extraction method of power signatures for Nonintrusive Load Monitoring Systems

The traditional Nonintrusive Load Monitoring Systems (NILMs) often employ average power consumption ex. real power and reactive power to recognize electric loads ON/OFF status. However, aiming at variable loads owning variable property in their power, voltage, and frequency, the unsteady power signatures will increase the difficulty degree of load identification. A new extraction method of power signatures is proposed in this paper, convolution and wavelet multi-resolution analysis (WMRA) technique are employed to extract the new power signature from the raw instantaneous power waveforms. Parseval´s theorem is employed to obtain the power indices (PIs) from the new power signature. After inputting PIs of each electric load, inner product is used to identify the unknown loads.