Applying Support Vector Machines and Boosting to a Non-Intrusive Monitoring System for Household Electric Appliances with Inverters

摘要

A non-intrusive load monitoring system has been developed for estimating the behavior of individual electrical appliances from the measurement of the total household load demand curve. The system is useful for monitoring both inverter and non-inverter type appliances that change their mode of operation over time. The total load demand is measured at the entrance of the feeder line into the house and the operating status of household electric appliances can be identified with the help of Support Vector Machines (SVM), Boosting, RBF and neural network techniques by analyzing the characteristic frequency content from the load curve of the household. Load curve measurements of air-conditioners, refrigerators (inverter type and non-inverter type), incandescent light, fluorescence light and television set are used as examples for training and test data. So far only a small data set was measured for this feasibility study and our experiments show a great potential for machine learning techniques. In particular the Boosting algorithm exhibits accurate classification of the operating status both for inverter and non-inverter type electric appliances.