NILM Algorithms General Comparison and Test of Adaptability
Tutor / Supervisor
Student
Pasquet, Arthur
Document type
Master thesis
Date
2020
rights
Open Access
Publisher
Universitat Politècnica de Catalunya
UPCommons
Abstract
NILM field is a hot spot in university and companies research due to the great advantages it
provides and its importance to reduce energy consumption within the households particularly.
This thesis allows a comparison between Benchmark and State-of-Art algorithms over various
datasets from different domains and measured by 12 metrics. It shows that the efficiency of an
algorithm depends very much on the metric used to measure it.
As a result, it is observed that algorithms using Deep Learning are generally superior to the
others, however it is not easy to rank them. The Transfer Learning tried between European
datasets underlines an encouraging lead, but on the contrary between American dataset it seems
unproductive.
This thesis carries out also the first multi-source Transfer Learning in the NILM field, concluding
the need of further experimentation to prove its relevancy
