Change search
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf
Monitoring Water Distribution Network using Machine Learning
KTH, School of Electrical Engineering and Computer Science (EECS), Network and Systems engineering.
2017 (English)Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesis
Abstract [en]

Water is an important natural resource. It is supplied to our home by water distribution network thatis owned and maintained by water utility companies. Around one third of water utilities across the globereport a loss of 40% of clean water due to leakage. The increase in pumping, treatment and operationalcosts are pushing water utilities to combat water loss by developing methods to detect, locate, and xleaks. However, traditional pipeline leakage detection methods require periodical inspection with humaninvolvement, which makes it slow and inecient for leakage detection in a timely manner. An alternativeis on-line, continuous, real-time monitoring of the network facilitating early detection and localization ofthese leakages. This thesis aims to nd such an alternative using various Machine Learning techniques.For a water distribution network, a novel algorithm is proposed based on the concept of dominantnodes from graph theory. The algorithm nds the number of sensors needed and their correspondinglocations in the network. The network is then sub-divided into several leakage zones, which serves as abasis for leak localization in the network. Thereafter, leakages are simulated in the network virtually,using hydraulic simulation software. The obtained time series pressure data from the sensor nodes ispre-processed using one-dimensional wavelet series decomposition by using daubechies wavelet to extractfeatures from the data. It is proposed to use this feature extraction procedure at every sensor nodelocally, which reduces the transmitted data to the central hub over the cloud thereby reducing the energyconsumption for the IoT sensor in real world.For water leakage detection and localization, a procedure for obtaining training data is proposed,which serves as a basis for recognition of patterns and regularities in the data using supervised Machinelearning techniques such as Logistic Regression, Support Vector Machine, and Articial Neural Network.Furthermore, ensemble of these trained model is used to build a better model for leakage detection andits localization. In addition, Random Forest algorithm is trained and its performance is compared tothe obtained ensemble of earlier models. Also, leak size estimation is performed using Support VectorRegression algorithm.It is observed that the sensor node placement using proposed algorithm provides a better leakage localizationresolution than random deployment of sensor. Furthermore, it is found that leak size estimationusing Support Vector Regression algorithm provides a reasonable accuracy. Also, it is noticed that RandomForest algorithm performs better than the ensemble model except for the low leakage scenario. Thus,it is concluded to estimate the leak size rst, based on this estimation for small leakage case ensemblemodels can be applied while for large leakage case only Random Forest can be used.

Abstract [sv]

Vatten ar en viktig naturresurs. Den levereras till vart hem via vattendistributionsnatet, som ags och underhalls av vattenforetag. Omkring en tredjedel av vattenforetagen over hela varlden rapporterar en forlustpa 40 % rent vatten pa grund av lackage. Okningen av pumpnings-, behandlings- och driftskostnader drivervattenforsorjningen till att bekampa vattenforluster genom att utveckla metoder for att upptacka, lokaliseraoch xa lackor. Emellertid kraver traditionella pipeline-detekteringsmetoder periodisk inspektion medstor skala mansklig inblandning, vilket gor det langsamt och ineektivt for lackage-detektion i tid. Ettalternativ ar on-line, kontinuerlig, realtidsovervakning av natverket som underlattar tidig detektering ochlokalisering av dessa lackage. Avhandlingen syftar till att hitta ett sadant alternativ med hjalp av olikamaskinlasningstekniker.For ett vattendistributionsnat foreslas en ny algoritm baserad pa begreppet dominerande noder frangrafteori. Algoritmen nner ut hur manga sensorer som behovs och deras motsvarande platser i natverket.Natverket delas sedan in i era lackagezoner, som utgor grunden for lackageplacering i natverket. Dareftersimuleras lackage i natverket praktiskt taget med hjalp av hydraulisk simuleringsprogramvara. Denerhallna tidsserie-tryckdatan fran sensornoderna forbehandlas med anvandning av endimensionell waveletseriebrytning genom att anvanda Daubechies Wavelet for att extrahera sardrag fran data. Det foreslas attanvanda detta extraktionsprocedur vid varje sensornod lokalt vilket minskar overford data till det centralanavet over molnet och darigenom minskar energiforbrukningen for IoT-sensorn i verkliga varlden.For upptackt och lokalisering av vattenlackage foreslas ett forfarande for erhallande av traningsdata,som utgor grunden for erkannande av monster och regelbundenhet i data som anvander overvakade maskininlarningstekniker, sasom logistik regression, stodvektormaskin och konstgjort neuralt natverk. Dessutomanvands ensemble av dessa tranad modeller for att bygga en battre modell for lackagespecikationoch lokalisering. Utover det ar Random Forest-algoritmen tranad och dess prestanda jamfordes med deterhallna ensemblet av tidigare modeller. Ocksa utmatning av lackstorlek utfors med hjalp av SupportVector Regression-algoritmen.Det observeras att sensorns nodplacering med anvandning av den foreslagna algoritmen ger en battrelackage-lokaliseringsupplosning an slumpmassig utplacering av sensorn. Vidare konstateras att lackstorleksuppskattningmed hjalp av supportvektorregressionsalgoritmen ger en rimlig noggrannhet. Det noterasocksa att Random Forest-algoritmen fungerar battre an ensemblemodellen med undantag for lag lackagescenario. Slutligen innebar detta att man uppskattar lackagestorleken forst. Baserat pa denna uppskattningfor sma lackagefall, kan ensemblemodeller appliceras medan for stort lackagefall kan endast RandomForest anvandas.

Place, publisher, year, edition, pages
2017. , p. 52
Series
TRITA-EE, ISSN 1653-5146 ; 2017:163
National Category
Electrical Engineering, Electronic Engineering, Information Engineering
Identifiers
URN: urn:nbn:se:kth:diva-221832OAI: oai:DiVA.org:kth-221832DiVA, id: diva2:1177842
Educational program
Master of Science - Wireless Systems
Available from: 2018-01-26 Created: 2018-01-26 Last updated: 2018-01-26Bibliographically approved

Open Access in DiVA

fulltext(1972 kB)244 downloads
File information
File name FULLTEXT01.pdfFile size 1972 kBChecksum SHA-512
9637b2b401319715aec887e5944544a2b342b1c3f34b56710398fac90a19d7321b6c81e7340c160962e9475729f1f212aaaeccfad575023ef6fc25a0810023c0
Type fulltextMimetype application/pdf

By organisation
Network and Systems engineering
Electrical Engineering, Electronic Engineering, Information Engineering

Search outside of DiVA

GoogleGoogle Scholar
Total: 244 downloads
The number of downloads is the sum of all downloads of full texts. It may include eg previous versions that are now no longer available

urn-nbn

Altmetric score

urn-nbn
Total: 216 hits
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf