Non-intrusive load monitoring by comparison of event probability density maps

摘要

A utility consumption event is identified by generating a consumption profile from a series of consumption values 1 (e.g. real and/or reactive power); detecting 3-8 an event there-within; comparing it 9 to a stored event profile 10 in the form of a probability density map; and identifying the detected event when they match. The probability that the detected event is of the type represented by the probability density map may be calculated (e.g. based on Mahalanobis distances) and a match determined when the probability exceeds a threshold; or for the highest probability. Events may be classified as non-periodic or periodic and compared to two- or three-dimensional probability density maps accordingly. Stored event profiles 10, (33, Figure 12A) may be populated by clustering events detected in an earlier consumption profile (21-22, Figure 12A) to produce a probability density map and comparing it to stored appliance probability density maps (24, Figure 12A).