Application of parallel computing in data mining for contaminant source identification in water distribution systems

Abstract

Contaminant source identification (CSI) procedures are drawing increasing attention due to the possibility of accidental and/or deliberate contaminant intrusion into water distribution systems. However, uncertainties that exist in the modeling have the potential to dramatically impact the capabilities of CSI procedures. Nodal demand uncertainties, as they influence false negative and false positive rates of contaminant detection, are examined. A procedure to quantify the false negative rate is provided, and the false positive issue is shown to be related to a parameter 'm'. Addressing the false positive and negative issues is demonstrated as feasible due to the use of parallel computing in a super-computer, which reduces the elapsed time for 150 scenario simulations from 37.5 hrs to only 15 min in the case study. By increasing the number of scenarios in the database for CSI through the use of a super-computer, the opportunity exists to decrease the false negative rate and reduce the number of false possible intrusion nodes.Keywords: parallel computing; contaminant source identification; uncertainty; EPANET; false positive; false negative Les procédures de l'identification des sources de contamination (CSI) sont en attirant l'attention croissante en raison de la possibilité de contact accidentel et/ou délibéré d'intrusion des contaminants dans les systèmes de distribution d'eau. Donc, les incertitudes qui existent dans la modélisation ont la potentialité de influer considérablement sur les capacités de la procédure de CSI. L'incertitude de la demande indispensables, car elles influencent les faux négatifs et de faux positifs des taux de détection des contaminants, qui sont examinés ici. Une procédure de quantification du taux des faux négatifs est préparée. Ainsi, la question de faux positifs est démontrée être liée à un paramètre 'm'.Se référant aux questions de faux positifs et négatifs, il est démontré comme étant dû à l'utilisation possible du calcul parallèle à l'aide d'un supercalculateur. Ce qui réduit le temps écoulé pour 150 simulations de scenarios de 37,5 heures à seulement 15 minutes dans cette étude. Il existe la possibilité pour diminuer ce taux de faux négatifs en augmentant le nombre de scénarios dans la base de donnée pour le CSI avec l'utilisation d'un supercalculateur et de réduire le nombre des faux noeuds d'intrusion possible.

RETRACTED: Mixed-mode database miner classifier: Parallel computation of graphical processing unit mining

“…17 In 2017, Kalaiselvi T, has applied GPU, CUDA computing in the medical image analysis and discuss about performances of existing algorithms are analyses and the computational gain is discussed. 18,19 In 2012, Ghorpade J reviewed about CUDA and its architecture and shows a comparison of CUDA C/Cþþ with other parallel programming languages such as OpenCL and Direct Compute. 20 In 2011, CUDA application design and development book were written by Farber Rob.…”

An Introduction to the special issue on tackling challenging water resources problems in Canada: a systems approach

“…In fact, all of the papers in this special issue are connected with one or more water topics written within the seven ellipses surrounding water in the bottom portion of Figure 1. In the next three papers, computational intelligence techniques are utilized to handle water resources issues (Asnaashari et al 2013;Shen and McBean 2013;Kamali et al 2013). With respect to a technological systems problem, Asnaashari et al (2013) forecast watermain failures using neural network modeling for use in water distribution and infrastructure rehabilitation and planning.…”

“…With respect to a technological systems problem, Asnaashari et al (2013) forecast watermain failures using neural network modeling for use in water distribution and infrastructure rehabilitation and planning. Shen and McBean (2013) employ parallel computing in data mining for identification of contaminant sources in water distribution systems. Then, Kamali et al (2013) use several heuristic approaches to calibrate hydrological models which are employed to predict river flows in the Smoky River watershed in Alberta.…”

A Spark-based genetic algorithm for sensor placement in large scale drinking water distribution systems

“…Wu et al [14] investigated the use of cloud computing in WDSs, where a pump scheduler has been deployed onto the high performance computer, through which a user can submit, execute and retrieve optimization analysis jobs. Shen et al [15] applied parallel computing to simulate intrusion events for CSI problem with a super-computer. Wang et al [16] presented a parallel method for the first time using the MapReduce paradigm to identify the contamination source in WDS.…”