Localizing pipe leaks is a significant challenge for water utilities worldwide. Pipe leaks in water distribution systems (WDSs) can cause the loss of a large amount of treated water, leading to pressure loss, increased energy costs, and contamination risks. What makes localizing pipe leaks challenging is the underground location of the water pipes and the similarity in impact on hydraulic properties (e.g., pressure, flow) due to leaks as compared to the effects of WDS operational changes. Physical methods to locate leaks are expensive, intrusive, and heavily localized. Computational approaches such as data-driven machine learning models provide an economical alternative to physical methods. Machine learning models are readily available and easily customizable to most problems; therefore, there is an increasing trend in their application for leak localization in WDSs. While several studies have applied machine learning models to localize leaks in single pipes and small test networks, these studies have yet to thoroughly test these models against the different complexities of leak localization problems, and hence their applicability to real-world WDSs is still unclear. The simplicity of the WDSs, the oversimplification of leak characteristics, and the lack of consideration of modeling and measuring device uncertainties adopted in most of these studies make the scalability of their proposed approaches questionable to real-world WDSs. Our study addresses this issue by devising four study cases of different complexity that account for realistic leak characteristics and model- and measuring device-related uncertainties. Two established machine learning models-multilayer perceptron (MLP) and convolutional neural network (CNN)-are trained and tested for their ability to localize the leaks and predict their sizes for each of the four study cases using different simulated hydraulic inputs. In addition, the potential benefit of combining different types of hydraulic data as inputs to the machine learning models in localizing leaks is also explored. Pressure and flow, two common hydraulic measurements, are used as inputs to the machine learning models. Further, the impact of single and multiple time point input in leak localization is also investigated. The results for the L-Town network indicate good accuracies for both the models for all study cases, with CNN consistently outperforming MLP.

Water distribution systems (WDSs) face a significant challenge in the form of pipe leaks. Pipe leaks can cause loss of a large amount of treated water, leading to pressure loss, increased energy costs, and contamination risks. Locating pipe leaks has been a constant challenge for water utilities and stakeholders due to the underground location of the pipes. Physical methods to detect leaks are expensive, intrusive, and heavily localized. Computational approaches provide an economical alternative to physical methods. Data-driven machine learning-based computational approaches have garnered growing interest in recent years to address the challenge of detecting pipe leaks in WDSs. While several studies have applied machine learning models for leak detection on single pipes and small test networks, their applicability to the real-world WDSs is unclear. Most of these studies simplify the leak characteristics and ignore modeling and measuring device uncertainties, which makes the scalability of their approaches questionable to real-world WDSs. Our study addresses this issue by devising four study cases that account for the realistic leak characteristics (multiple, multi-size, and randomly located leaks) and incorporating noise in the input data to account for the model- and measuring device- related uncertainties. A machine learning-based approach that uses simulated pressure as input to predict both location and size of leaks is proposed. Two different machine learning models: Multilayer Perceptron (MLP) and Convolutional Neural Network (CNN), are trained and tested for the four study cases, and their performances are compared. The precision and recall results for the L-Town network indicate good accuracies for both the models for all study cases, with CNN generally outperforming MLP.

A multi-reservoir simulation-optimization model GRAPS, Generalized Multi-Reservoir Analyses using Probabilistic Streamflow Forecasts, is developed in which reservoirs and users across the basin are represented using a node-link representation. Unlike existing reservoir modeling software, GRAPS can handle probabilistic streamflow forecasts represented as ensembles for performing multi-reservoir prognostic water allocation and evaluate the reliability of forecast-based allocation with observed streamflow. GRAPS is applied to four linked reservoirs in the Jaguaribe Metropolitan Hydro-System (JMH) in Ceará, North East Brazil. Results from the historical simulation and the zero-inflow policy over the JMH system demonstrate the model's capability to support monthly water allocation and reproduce the observed monthly releases and storages. Additional analyses using streamflow forecast ensembles illustrate GRAP's abilities in developing storage-reliability curves under inflow-forecast uncertainty. Our analyses show that GRAPS is versatile and can be applied for 1) short-term operating policy studies, 2) long-term basin-wide planning evaluations, and 3) climate-information based application studies.

Fluid pressure influences leakage flow rate in water distribution pipe networks. Significant progress has been made in the use of pressure management techniques to control leakage. An empirical power equation (known as the N1 power equation) is widely used to model the pressure-leakage relationship in practice. However, recent research has shown that this equation is not able to accurately model leakage under various conditions. When estimating the N1 leakage exponent for a system, the power equation assumes that leakage is located at a point in the network that is representative of the average system pressure, which is rarely true. In addition, parameters of the power equation are not constant but vary with pressure, which is a problem especially for water distribution pipe networks where pressure varies with time, for example, due to diurnal demand variation. These factors result in significant errors when the power equation is used to model leakage in real water distribution networks. This technical note analyzes the factors that introduce errors in the power equation and proposes methods to improve its accuracy. The performance of the methods is demonstrated using a case study.

The long-term effect of non-point source pollution on groundwater from agricultural practices is a major concern globally. Non-point source pollutants such as nitrate that occurs through fertilizers and animal waste eventually make their way into the aquifer by infiltrating soil. The goal of this study is to develop an approach for characterization of nitrate concentrations at potential source locations under conditions of geologic uncertainty. A Bayesian framework using the Markov Chain Monte Carlo (MCMC) approach is developed to estimate posterior probability distributions of non-point sources by incorporating nitrate concentration data as well as geologic uncertainties. The proposed approach is tested using hypothetical contamination scenarios and then validated using an application case study in North Carolina. Uncertainty existing in geologic formation (i.e., heterogeneous hydraulic conductivity field) is treated as prior and used in evaluating the likelihood function that measures the match between observed and simulated concentrations. The likelihood function computation involves a numerical model that simulates nitrate transport in groundwater from non-point agricultural sources and predicts nitrate concentrations at observation wells. Effectiveness of the MCMC approach is evaluated through a convergence analysis. Comparison among different sampling algorithms is carried out with respect to MCMC convergence diagnostics and making inference. The Bayesian inference analysis methodology developed in this research will help decision makers and water managers to identify potential areas for source containment and decide if further sampling is required.

In many modern water networks, an emerging trend is to measure pressure at various points in the network for operational reasons. Because leaks typically induce a signature on pressure, these routine measurements can be used to develop nonintrusive leak detection approaches. This research employs successive linear approximation methods, based on linear programming and mixed integer linear programming, in a simulation-optimization framework to explore an alternative leak detection methodology for urban water distribution networks based on pressure measurements. The methods attempt to minimize the absolute differences between observed and simulated pressure values at the sensors to determine a linear combination of leaks that most closely approximates the observed pressure pattern. Steady-state and time-varying models of differing complexity (from small published networks to a 27,000-node network for a U.S. utility) were used to test the method. Results are presented to illustrate the method's effectiveness under different conditions. The methods are shown to work well when pervasive pressure data and hydraulic models representing true operational conditions are available. The methods developed in this work are not intended to replace traditional leak detection methods; rather, they are meant to work in concert with available methods to more accurately and efficiently isolate leak locations and reduce water loss. (C) 2017 American Society of Civil Engineers.

Municipal water demands in growing population centers in the arid southwest US are typically met through increased groundwater withdrawals. Hydro-climatic uncertainties attributed to climate change and land use conversions may also alter demands and impact the replenishment of groundwater supply. Groundwater aquifers are not necessarily confined within municipal and management boundaries, and multiple diverse agencies may manage a shared resource in a decentralized approach, based on individual concerns and resources. The interactions among water managers, consumers, and the environment influence the performance of local management strategies and regional groundwater resources. This research couples an agentbased modeling (ABM) framework and a groundwater model to analyze the effects of different management approaches on shared groundwater resources. The ABM captures the dynamic interactions between household-level consumers and policy makers to simulate water demands under climate change and population growth uncertainties. The groundwater model is used to analyze the relative effects of management approaches on reducing demands and replenishing groundwater resources. The framework is applied for municipalities located in the Verde River Basin, Arizona that withdraw groundwater from the Verde Formation-Basin Fill-Carbonate aquifer system. Insights gained through this simulation study can be used to guide groundwater policy-making under changing hydro-climatic scenarios for a long-term planning horizon.

Finding the location and concentration of contaminant sources is an important step in groundwater remediation and management. This discovery typically requires the solution of an inverse problem. This inverse problem can be formulated as an optimization problem where the objective function is the sum of the square of the errors between the observed and predicted values of contaminant concentration at the observation wells. Studies show that the source identification accuracy is dependent on the observation locations (i.e., network geometry) and frequency of sampling; thus, finding a set of optimal monitoring well locations is very important for characterizing the source. The objective of this study is to propose a sensitivity-based method for optimal placement of monitoring wells by incorporating two uncertainties: the source location and hydraulic conductivity. An optimality metric called D-optimality in combination with a distance metric, which tends to make monitoring locations as far apart from each other as possible, is developed for finding optimal monitoring well locations for source identification. To address uncertainty in hydraulic conductivity, an integration method of multiple well designs is proposed based on multiple hydraulic conductivity realizations. Genetic algorithm is used as a search technique for this discrete combinatorial optimization problem. This procedure was applied to a hypothetical problem based on the well-known Borden Site data in Canada. The results show that the criterion-based selection proposed in this paper provides improved source identification performance when compared to uniformly distributed placement of wells.

It is estimated that 15-40% of water is unaccounted for in urban water systems. This is mostly caused by small leaks, which are difficult to locate. Routinely measured pressure, flow, and water quality data can be used to locate leaks in the water network using an inverse modeling approach. For a known sensor configuration, the leak locations can be found by minimizing the difference between real and simulated measurements. However, when comparing measurement types (pressure, flow, or quality), some may be more sensitive to leak location than others. Furthermore, some measurement types may be more or less sensitive depending on the leak magnitude or the proximity of the leak to the sensors. The measurements types that are more sensitive to location will have a stronger signature and would need to be weighted more in an inverse modeling approach, especially in the presence of noise. Preliminary research suggests that water quality measurements are more sensitive to leak location when a leak is small, and that flow measurements are more sensitive when a leak is large. In this research, a series of sensitivity analyses are conducted on different networks to investigate the sensitivity of these measurements with respect to leak location, magnitude, and proximity of sensors to the leak location.

Inefficient parallel I/O is known to be a major bottleneck among scientific applications employed on supercomputers as the number of processor cores grows into the thousands. Our prior experience indicated that parallel I/O libraries such as HDF5 that rely on MPI-IO do not scale well beyond 10K processor cores, especially on parallel file systems (like Lustre) with single point of resource contention. Our previous optimization efforts for a massively parallel multi-phase and multi-component subsurface simulator (PFLOTRAN) led to a two-phase I/O approach at the application level where a set of designated processes participate in the I/O process by splitting the I/O operation into a communication phase and a disk I/O phase. The designated I/O processes are created by splitting the MPI global communicator into multiple sub-communicators. The root process in each sub-communicator is responsible for performing the I/O operations for the entire group and then distributing the data to rest of the group. This approach resulted in over 25X speedup in HDF I/O read performance and 3X speedup in write performance for PFLOTRAN at over 100K processor cores on the ORNL Jaguar supercomputer. This research describes the design and development of a general purpose parallel I/O library called Scorpio that incorporates our optimized two-phase I/O approach. The library provides a simplified higher level abstraction to the user, sitting atop existing parallel I/O libraries (such as HDF5) and implements optimized I/O access patterns that can scale on larger number of processors. Performance results with standard benchmark problems and PFLOTRAN indicate that our library is able to maintain the same speedups as before with the added flexibility of being applicable to a wider range of I/O intensive applications.

Leak detection and management is an important problem in water distribution systems because it has been documented that up to 40% of the water may be lost to leaks in many aging systems. Small gradual leaks, which represent more than half of all leaks, are difficult to locate. Routinely measured pressure, flow, and water quality data in combination with a simulation-optimization inverse modeling approach could be used to characterize leakage. In this approach, the leak locations are found by minimizing the difference between real and simulated measurements for a known sensor configuration. Simulation-optimization approaches are computationally demanding because millions of simulations of a network simulator (e.g., EPANET) may be required to achieve a satisfactory solution. This problem is alleviated using a high-performance computing (HPC) framework that enables many parallel simulations of the water system using EPANET. This research is modifying an existing global search algorithm, called the Dividing Rectangles (DIRECT) Search that is traditionally used for continuous functions to enable parallel simulations and a mix of discrete variables (for leak locations) and continuous variables (for leak magnitudes). The modified algorithm is being tested with traditional continuous test functions, discrete test functions, and test water distribution networks.

Parallel computing has become a powerful approach for solving real-time decisions about large-scale, computing-intensive transportation problems. A frequently encountered transportation problem is the "shortest path problem;" that is, finding the shortest path between any two nodes in a transportation network. For the large transportation networks encountered in major metropolitan areas, this problem can be computationally demanding, especially if shortest paths between all the nodes in the network need to be dynamically updated (e. g., evolving traffic conditions). In such a situation, one may wish to harness parallel computing to solve this problem. However, the parallel implementations of commonly used shortest-path algorithms are computationally demanding because of the inherent sequential nature of the search process used by the algorithms. This paper describes parallel implementations and includes performance analyses of two prominent graph algorithms (i.e., Floyd-Warshall and Dijkstra) used for finding the all-pairs shortest path for a large-scale transportation network. The results indicate that a multilevel parallel implementation that combines message passing interface (MPI) with shared memory threads [e. g., Open Multiprocessing (OpenMP) or POSIX Threads (pthreads)] is effective for solving these problems on a moderate number of symmetric multiprocessor (SMP) nodes. This paper also includes the derivation of the computational time for the different parallel implementations of these two graph algorithms. DOI: 10.1061/(ASCE)CP.1943-5487.0000220. (C) 2013 American Society of Civil Engineers.

Population based heuristic search methods such as evolutionary algorithms (EA) and particle swarm optimization (PSO) methods are widely used for solving optimization problems especially when classical techniques are inadequate. A parallel optimization framework using multiple concurrent particle swarms is developed and applied to water distribution problems. Details of the enabling framework that couples the optimization methods with a parallel simulator built around EPANET will be discussed. In addition, algorithmic and computational performance results using ORNL's and ANL's leadership class parallel architectures will be presented for leakage detection and contaminant source characterization problems for two water distribution networks with 1,834 and 12,457 nodes respectively.

This study investigates and discusses groundwater system characterization problem utilizing surrogate modeling. In this inverse problem, the contaminant signals at monitoring wells are recorded to recreate the pollution profiles. In this study, simulation-optimization approach is a technique utilized to solve inverse problems by formulating them as an optimization model, where evolutionary computation algorithms are used to perform the search. In this approach, the partial differential equations (PDE) groundwater transport simulation model is solved iteratively during the evolutionary search, which in general can be computationally expensive since thousands of simulation model evaluations will be evaluated. To overcome this limitation, the simulation model is replaced by a surrogate model, which is computationally much faster than the simulation model and yet is relatively accurate. Artificial neural networks (ANN) is used to construct surrogate models that provide acceptable accuracy performances. The ANN surrogate model, which replaces the PDE groundwater transport simulation model, is then coupled with a genetic algorithm (GA) search procedure to solve the source identification problem. The results will present the quality solution of the ANN surrogate model versus the groundwater simulation model, the solution of the inverse problem for different experiment scenarios and finally a timing study analysis conducted to measure the surrogate model performance.

Water Distribution Systems are one of the most substantial and vulnerable part of civil infrastructure systems. For the reason that many large water distribution systems are old, which results in more leakage and expenses (e.g., increasing pump head, pipe burst, constituents’ replacement), a significant portion of water produced by the utilities never passes through the consumers’ meters. Due to the complex nature and vast spatial extent of a water distribution system it may be difficult for the utility personnel to identify and fix the leaks, therefore it is imperative to develop software frameworks for modeling and analyzing leakage in water distribution system during ordinary operational conditions as well as unexpected events. In this paper a Bayesian approach with Markov chain Monte Carlo method is implemented to map probabilistic characterizations of water leakage. If for this purpose physical parameters such as pipe vintage, material, and loading are available, they can be are used to develop prior information; otherwise, a uniform prior may be assumed. Routinely measured water quality, pressure, and flow measurements together with the uncertainty in demand are used to develop the likelihood function. The analyses are facilitated through the EPANET water distribution simulation tool. The efficiency and versatility of the proposed methodology is examined using water distribution network.

This poster describes the development of a highly scalable application layer parallel I/O library (ASCEM-IO) for scientific applications. This library was envisioned to leverage our earlier I/O optimization experience to build a scalable general purpose parallel I/O capability for any application. The parallel I/O library provides a higher level API (Application Programming Interface) to read and write large scientific datasets in parallel at very large processor counts. Specifically, the goal is to take advantage of existing parallel I/O libraries, such as HDF5 which are being widely used by scientific applications and modify these algorithms to better scale on larger number of processors. This is accomplished by dividing the traditional I/O operations (read/write) into two phases, a communication phase and an I/O phase. Results with a real application on the Cray XT/5 indicates significant performance improvement on large processor cores when compared to default HDF collective I/O operations.

We present a workflow-based algorithm for identifying threads to an urban water management system. Through Grid computing we provide the necessary high-performance computing resources to deliver quickly solutions to the problem. We prototyped a new middleware called cyberaide, that enables easy access to Grid resources through portals or the command line. A workflow system is used to manage resources in fault tolerant fashion. In addition, we contrast the architecture with a Hadoop implementation. Resources from TeraGrid and FutureGrid are used to test the feasibility of using the toolkit for a scientific application.

In response to the enormous and growing complexity of today’s high-end systems, the Performance Engineering Research Institute (PERI) is working toward automating portions of the performance tuning process by developing an autotuning framework. Our framework employs empirical techniques to identify the best-performing version of a computation among a search space of possible implementations. This paper describes application of PERI performance tools to demonstrate performance gains on production scientific codes on DOE’s leadership-class systems. This paper examines the use of PERI tools in automatic and semiautomatic tuning of PFLOTRAN, which models subsurface flow in groundwater.

Many engineering and environmental problems that involve the determination of unknown system characteristics from observation data can be categorized as inverse problems. A common approach undertaken to solve such problems is the simulation-optimization approach where simulation models are coupled with optimization or search methods. Simulation-optimization approaches, particularly in environmental characterization involving natural systems, are computationally expensive due to the complex three-dimensional simulation models required to represent these systems and the large number of such simulations involved. Emerging grid computing environments (e.g., TeraGrid) show promise for improving the computational tractability of these approaches. However, harnessing grid resources for most computational applications is a nontrivial problem due to the complex hierarchy of heterogeneous and geographically distributed resources involved in a grid. This paper reports and discusses the development and evaluation of a grid-enabled simulation-optimization framework for solving environmental characterization problems. The framework is designed in a modular fashion that simplifies coupling with simulation model executables, allowing application of simulation-optimization approaches across problem domains. The framework architecture utilizes standard communications protocols and the message passing interface with an application programming interface to establish a connection between a centralized search application and simulation models running on TeraGrid resources. Sets of performance and scalability results for solving a groundwater source history reconstruction (SHR) problem are presented. The results show that for a given set of resources, parameters controlling the granularity at various levels of parallelism play an important role in the overall parallel performance. A production run for solving the SHR problem using three geographically distributed grid resources indicates that even in a cross-site grid environment a factor of 90 speedup is possible using 140 computer processors.