Time-frequency characterisation for electric load monitoring

Electric utilities and consumers are increasingly interested in energy monitoring for economic and environmental reasons. A non-intrusive solution may rely on information extracted from the electric consumption measured at a centralized part of a distribution network. The problem at hands consists in the separation of the electric load into its major components. This problem of source separation from one sensor is quite tractable under certain conditions. In this work, the focus is made on the most consuming household appliance in France: the space-heating. It is a sum of an unknown number of pseudo-periodic signals embedded in the global active power. An unsupervised algorithm to determine the space-heating schedule from the global consumption based on the interpretation of the space-heating signature in the time-frequency domain is proposed. The proposed method conjoins a time-frequency detector and a frequent itemsets extraction. First results on real data are quite satisfying.