An Open-Source, Semisupervised Water End-Use Disaggregation and Classification Tool
Nour A. Attallah, Jeffery S. Horsburgh, Camilo J. Bastidas Pacheco
Abstract
This paper demonstrates a new water end-use disaggregation and classification tool that builds on existing end-use disaggregation studies and addresses the unavailability of code and data used by prior studies. The tool was developed in Python and can be accessed via any current Python programming environment. The base disaggregation and classification model for the tool was developed and tested on high-temporal-resolution data for a single home at which manually labeled end-use event data were also collected. The tool was then applied to four additional homes selected from a larger data set for 31 homes located in the cities of Logan and Providence, Utah, to demonstrate the generalizability of the tool. At homes for which no manually labeled end-use data are available, the tool’s base model is extended through a self-learning procedure that trains the model for an individual home using end-use events identified at that home. Results from homes with different meter types and sizes are presented to demonstrate the ability of the tool to disaggregate and classify high-temporal-resolution data into individual end-use events. The results of this paper are reproducible using openly available code and data, representing an accessible platform for advancing end-use disaggregation tools. The tool can be adapted to specific research needs.