Disaggregation by State Inference A Probabilistic Framework For Non-Intrusive Load Monitoring
Loading...
URL
Journal Title
Journal ISSN
Volume Title
Perustieteiden korkeakoulu |
Master's thesis
Unless otherwise stated, all rights belong to the author. You may download, display and print this publication for Your own personal use. Commercial use is prohibited.
Authors
Date
2016-01-18
Department
Major/Subject
Mcode
T61
Degree programme
Master’s Programme in Machine Learning and Data Mining (Macadamia)
Language
en
Pages
58
Series
Abstract
Non-intrusive load monitoring (NILM), the problem of disaggregating whole home power measurements into single-appliance measurements, has received increasing attention from the academic community because of its energy saving potentials, however the majority of NILM approaches are either variants of event-based or event-less disaggregation. Event-based approaches are able to capture much information about the transient behavior of appliances but suffer from error-propagation problems whereas event-less approaches are lessprone to error-propagation problems but can only incorporate transient information to a small degree. On top of that inference techniques for event-less approaches are either computationally expensive, do not allow to trade off computational time for approximation accuracy or are prone to local minima. This work will contribute three-fold: first an automated way to infer ground truth from single appliance readings is introduced, second an augmentation for event-less approaches is introduced that allows to capture side-channel as well as transient information of change-points, third an inference technique is presented that allows to control the trade-off between computational expense and accuracy. Ultimately, this work will try to put the NILM problem into a probabilistic framework that allows for closing feedback loops between the different stages of event-based NILM approaches, effectively bridging event-less and event-based approaches. The performance of the inference technique is evaluated on a synthetic data set and compared to state-of-the-art approaches. Then the hypothesis that incorporating transient information increases the disaggregation performance is tested on a real-life data set.Description
Supervisor
Karhunen, JuhaThesis advisor
Berges, MarioKeywords
hidden Markov models, energy disaggregation, load monitoring, efficient inference