Feature extraction for nonintrusive load monitoring based on S-Transform

The electric energy demand is dramatically growing worldwide and demand reduction emerges as an outstanding strategy; it implies detailed information about the electricity consumption, namely load disaggregation. Typical automatic methods for load disaggregation require high hardware efforts to install one sensor per appliance, whereas Non-intrusive Load Monitoring (NILM) systems diminish the hardware efforts through signal processing and mathematical modeling. One approach to NILM systems is to model the load signatures via artificial intelligence. This paper proposes to employ S-Transform for the feature extraction stage and Support Vector Machines for the pattern recognition problem. Several experiments are presented and the results of the feature extraction with S-Transform and Wavelet Packet Transform are compared. Thus promising feature vectors based on S-Transform are presented with similar or superior performance than the approach based on Wavelet Packet Transform.