Enhanced approaches for non-intrusive load disaggregation

• Main Author: Aiad, Misbah M. M.
• Other Authors: Lee Peng Hin
• Format: Theses and Dissertations
• Language: English
• Published: 2018
• Subjects: DRNTU::Engineering::Computer science and engineering::Computing methodologies::Simulation and modeling; DRNTU::Engineering::Mechanical engineering::Energy conservation
• Online Access: https://hdl.handle.net/10356/90196; http://hdl.handle.net/10220/47182
• Institution: Nanyang Technological University

The modern urban life and increasing demands of energy are calling toward energy conservation and energy efficient strategies. Energy saving and energy management in the residential sectors are of great interests for obvious economic and environmental reasons, with increasing energy consumptions by the consumers. An efficient energy conservation and monitoring program requires some means of monitoring the power consumed by individual appliances within the households. The deployment of smart meters in smart grids in many countries has generated an increase in research interests in the areas of non-intrusive load monitoring (NILM) in recent years. Non-intrusive load monitoring, or load disaggregation, are sets of techniques and methods that decompose the total aggregate consumptions, measured at a single point by smart meters, into the respective appliance-specific consumptions in the household. Studies conducted have shown that information of the energy consumed by individual appliances in the homes can influence the behavior of the household occupants in a way that can achieve noticeable energy savings. There are several challenges in the domain of unsupervised load disaggregation approaches that do not require human intervention for learning or installation of additional measuring instruments for each appliance, apart from the smart meters, allowing a feasible economic adoption of NILM techniques. In this thesis, a detailed literature review on methods and techniques applied to NILM and common challenges is presented. Enhanced approaches that tackle three essential challenges in the domain of NILM were proposed. Firstly, with the aim to achieve an improved disaggregation accuracy, an unsupervised approach for load disaggregation that embeds the mutual devices interactions information into the factorial hidden Markov model (FHMM) representation of the total aggregate signal was introduced. The method was further extended with adaptive estimations of the devices main power consumptions effects and their two-way interactions. Secondly, the modeling of continuously varying loads was proposed using a quantized continuous-state hidden Markov model (CS-HMM). A method to estimate the transition matrix that mitigates the both extreme cases of too frequent and never occurred transitions was introduced and the Viterbi algorithm was used to estimate the power consumption profile of the variable loads. Thereafter, the proposed model for the continuously varying loads was integrated with the standard FHMM to produce a hybrid continuous/discrete state HMM, which is capable of modeling and disaggregating energy consumptions from a wider range of home appliances types. Thirdly, to tackle the problem of overlapping clusters that represent devices power consumptions resulting when applying a clustering-based disaggregation, a method to analyze the cohesion of devices’ clusters to determine if a cluster should be split into two small clusters was proposed. The analysis of clusters cohesion was investigated based on normality tests performed against two confidence levels. The proposed approaches and techniques were applied and tested on real houses from the Reference Energy Disaggregation Data Set (REDD). The proposed approaches, in general, enhanced the overall performance and accuracy of disaggregation. The work presented in thesis represents an advancement in the state-of-art in the domain of NILM and contributes toward achieving energy savings in residential homes.