Water use disaggregation based on classification of feature vectors extracted from smart meter data

M. Vašak, G. Banjac and H. Novak

in Computing and Control for the Water Industry Conference, Leicester, UK, September 2015.
BibTeX  URL 

@inproceedings{VBN:2015,
  author = {M. Va\v{s}ak and G. Banjac and H. Novak},
  title = {Water use disaggregation based on classification of feature vectors extracted from smart meter data},
  booktitle = {Computing and Control for the Water Industry Conference},
  year = {2015},
  url = {http://dx.doi.org/10.1016/j.proeng.2015.08.992},
  doi = {10.1016/j.proeng.2015.08.992}
}

This work studies a feasibility of water use disaggregation from commercial smart meters, with volume quantization on the level of 1 litre. New water use characterisation is performed by extraction of predefined features from smart meter records and determination of a region from a given pre-calculated region partition to which the corresponding features belong. Region adherence corresponds to a particular fixture type actuation in the household. Regions partition is computed by employing multicategory robust linear programming technique. Performance of the water use disaggregation model is shown when a separate model is tuned for each household based on sole household data.