Show simple item record

dc.contributor.advisorZhang, Hong
dc.contributor.authorYang, Ao
dc.date.accessioned2019-02-20T23:41:49Z
dc.date.available2019-02-20T23:41:49Z
dc.date.issued2018-10
dc.identifier.doi10.25904/1912/2660
dc.identifier.urihttp://hdl.handle.net/10072/382672
dc.description.abstractIn the urban water planning and management industry, end-use water consumption monitoring is a primary tool for water demand management and source substitution. Numerous residential end-use consumption studies have been carried out worldwide in the last two decades. With the rapid development of intelligent technology, the traditional time-consuming process for water flow data disaggregation has been replaced by a smart water metering system with advanced analysis. However, the existing water flow trace analysis system cannot accurately disaggregate all categories of residential water end-use events. In response to this issue, this research focused on developing new techniques, which can improve the autonomous categorisation accuracy of the residential water flow disaggregation process. A rigorous research method was adopted to achieve the above-mentioned research objectives and included the following two stages: (1) review and testing of pattern recognition techniques; and (2) software development. This study employed the extensive South-east Queensland (SEQ) Residential Water End Use Study dataset to undertake the development of the intelligent and autonomous water end-use recognition technique. Due to the array of objectives, methods, and results, this thesis has been structured around two refereed journal publications produced during the MPhil study. Two themes emerged from the research, namely: (1) development of hybrid intelligent model for mechanised water end-use analysis; and (2) optimising water end-use analysis process with Self-organising maps and K-Means clustering. The application of many sophisticated intelligent techniques has been attempted in order to tackle this complex problem. In the first stage, the original application of Dynamic Time Warping (DTW) algorithm was found to be ineffective due to settings of the threshold value. Through further investigation into the existing database, Artificial Bee Colony (ABC) and K-Medoids algorithm were selected. In this stage, this technique was applied to assist in finding toilet events in an artificially mixed data. 95.71% accuracy for correctly classified mechanical events was achieved when tested on 136 mixed events from different categories. The performance of the selected algorithms have been compared against previously reported approaches, with the technique and accuracy comparisons presented in a refereed journal paper. While the ABC and K-Medoids approach to clustering flow data into water end-use categories was suitable for mechanical end-use categories, it was less effective for other behaviourally influenced categories. Further exploration of various water flow data clustering techniques was required in order to discover a more suitable approach for the preliminary clustering of flow data into all of the water end-use categories. This prompted the undertaking of the research activities for the second journal paper described as follows. The study continued with the development of a hybrid technique in the second stage. Self-organising maps (SOM) and K-means algorithms were applied to the existing software Autoflow through pre-grouping of water end-use events in order to improve the accuracy. The verification on two datasets (i.e., (1) over 100,000 single events, and (2) 30 independent homes), resulted in an improvement in water end-use categorisation accuracy, when compared to the original technique employed in Autoflow, for each residential end-use category. Accuracy improvements were particularly noticeable for the mechanical water end-use event categories (i.e., washing machine, toilet, and evaporative cooler). The research outcomes have implications for researchers and the water industry. For researchers, the revised Autoflow v3.1 developed in this study is more accurate than previous versions reported in the literature. The novel hybrid pattern recognition approach and the associated algorithms employed in this latest Autoflow v3.1 version can be adapted for a range of pattern recognition problems. For the water industry, an accurate and autonomous water end-use analysis software tool has a range of implications, including, providing bottom-up data for demand forecasting and infrastructure planning, evidence-based water demand management, and end-use level customer feedback phone and web-based applications.
dc.languageEnglish
dc.language.isoen
dc.publisherGriffith University
dc.publisher.placeBrisbane
dc.subject.keywordsArtificial intelligent techniques
dc.subject.keywordsResidential water
dc.subject.keywordsUrban water management
dc.subject.keywordsSoftware development
dc.subject.keywordsReview testing
dc.titleArtificial Intelligent Techniques in Residential Water End-use Studies for Optimized Urban Water Management
dc.typeGriffith thesis
gro.facultyScience, Environment, Engineering and Technology
gro.rights.copyrightThe author owns the copyright in this thesis, unless stated otherwise.
gro.hasfulltextFull Text
dc.contributor.otheradvisorStewart, Rodney
dc.contributor.otheradvisorNguyen, Khoi
gro.thesis.degreelevelThesis (Masters)
gro.thesis.degreeprogramMaster of Philosophy (MPhil)
gro.departmentSchool of Eng & Built Env
gro.griffith.authorYang, Ao


Files in this item

This item appears in the following Collection(s)

Show simple item record