Contaminant Source Localization Research Articles

Bayesian optimization with informative parametric models via sequential Monte Carlo

AbstractBayesian optimization (BO) has been a successful approach to optimize expensive functions whose prior knowledge can be specified by means of a probabilistic model. Due to their expressiveness and tractable closed-form predictive distributions, Gaussian process (GP) surrogate models have been the default go-to choice when deriving BO frameworks. However, as nonparametric models, GPs offer very little in terms of interpretability and informative power when applied to model complex physical phenomena in scientific applications. In addition, the Gaussian assumption also limits the applicability of GPs to problems where the variables of interest may highly deviate from Gaussianity. In this article, we investigate an alternative modeling framework for BO which makes use of sequential Monte Carlo (SMC) to perform Bayesian inference with parametric models. We propose a BO algorithm to take advantage of SMC’s flexible posterior representations and provide methods to compensate for bias in the approximations and reduce particle degeneracy. Experimental results on simulated engineering applications in detecting water leaks and contaminant source localization are presented showing performance improvements over GP-based BO approaches.

Application of an iterative source localization strategy at a chlorinated solvent site

This study presents an inverse modeling strategy for organic contaminant source localization. The approach infers the hydraulic conductivity field, the dispersivity, and the source zone location. Beginning with initial observed data of contaminant concentration and hydraulic head, the method follows an iterative strategy of adding new observations and revising the source location estimate. Non-linear optimization using the Gauss-Levenberg-Marquardt Algorithm (PEST++) is tested at a real contaminated site. Then a limited number of drilling locations are added, with their positions guided by the Data Worth analysis capabilities of PYEMU. The first phase of PEST++, with PYEMU guidance, followed by addition of monitoring wells provided an initial source location and identified four additional drilling locations. The second phase confirmed the source location, but the estimated hydraulic conductivity field and the Darcy flux were too far from the measured values. The mismatch led to a revised conceptual site model that included two distinct zones, each with a plume emanating from a separate source. A third inverse modelling phase was conducted with the revised site conceptual model. Finally, the source location was compared to results from a Geoprobe@ MiHPT campaign and historical records, confirming both source locations. By merging measurement and modeling in a coupled, iterative framework, two contaminant sources were located through only two drilling campaigns while also reforming the conceptual model of the site.

Aggregated GP-based Optimization for Contaminant Source Localization

Recently a new simulation-based optimization benchmark of groundwater contaminant source localization problems has been introduced to the hydrogeological science community. Given information on contaminant concentration levels at each monitoring well and each time step, its objective is to identify the location of contaminant source. In this work, we analyze and look at the problem from different angles to gain more insights on this class of groundwater problems. To tackle the problem, a novel simulation-based optimization algorithm relying on an aggregated Gaussian process model, and the expected improvement criterion is introduced. Results from this study show that the proposed algorithm, though relying on an approximated Gaussian process model, demonstrates superior efficiency and reliability than a traditional expected improvement-based algorithm. The location of the monitoring wells was confirmed to play a crucial role in assisting the optimization algorithm to accurately localize the contaminant source. Additional monitoring wells, while adding more knowledge of the space-time mapping of concentration levels, could nevertheless slow down convergence of the algorithm due to the increase in problem complexity.

Application of the artificial neural network in the forecasting of the airborne contaminant

Release of hazardous materials in chemical industries is a significant threat to surrounding areas. This thread can be answered by the reconstruction system capable of localizing the source of airborne contamination solely based on substance concentrations recorded by the sensors network. However, such systems require multiple runs of the selected atmospheric contaminant transport model. The complexity of the contaminated terrain involves the application of the complicated and computationally expensive dispersion models, while the fast one is too simplified. We examine the possibility of training an artificial neural network (ANN) so that it could effectively simulate the atmospheric toxin transport. The use of a fast neural network in place of costly computational dispersion models in systems localizing the source of contamination might significantly improve their efficiency (speed). In this paper, we train the ANN with the use of the training dataset covering the contamination source term parameters and point output concentrations generated by the Gaussian dispersion model. We test various ANN structures, i.e., numbers of ANN layers, neurons, and activation functions to achieve the ANN capable of estimating the contaminant concentration. Applying the specified ANN topology we train ANN with use of the real field Prairie Grass experiment data. The performed tests confirm that trained ANN has the potential to replace the dispersion model in the contaminant source localization systems.

Contaminant source localization via Bayesian global optimization

Abstract. Contaminant source localization problems require efficient and robust methods that can account for geological heterogeneities and accommodate relatively small data sets of noisy observations. As realism commands hi-fidelity simulations, computation costs call for global optimization algorithms under parsimonious evaluation budgets. Bayesian optimization approaches are well adapted to such settings as they allow the exploration of parameter spaces in a principled way so as to iteratively locate the point(s) of global optimum while maintaining an approximation of the objective function with an instrumental quantification of prediction uncertainty. Here, we adapt a Bayesian optimization approach to localize a contaminant source in a discretized spatial domain. We thus demonstrate the potential of such a method for hydrogeological applications and also provide test cases for the optimization community. The localization problem is illustrated for cases where the geology is assumed to be perfectly known. Two 2-D synthetic cases that display sharp hydraulic conductivity contrasts and specific connectivity patterns are investigated. These cases generate highly nonlinear objective functions that present multiple local minima. A derivative-free global optimization algorithm relying on a Gaussian process model and on the expected improvement criterion is used to efficiently localize the point of minimum of the objective functions, which corresponds to the contaminant source location. Even though concentration measurements contain a significant level of proportional noise, the algorithm efficiently localizes the contaminant source location. The variations of the objective function are essentially driven by the geology, followed by the design of the monitoring well network. The data and scripts used to generate objective functions are shared to favor reproducible research. This contribution is important because the functions present multiple local minima and are inspired from a practical field application. Sharing these complex objective functions provides a source of test cases for global optimization benchmarks and should help with designing new and efficient methods to solve this type of problem.

Deep Gaussian Process-Based Bayesian Inference for Contaminant Source Localization

This paper proposes a Bayesian framework for localization of multiple sources in the event of accidental hazardous contaminant release. The framework assimilates sensor measurements of the contaminant concentration with an integrated multizone computational fluid dynamics (multizone-CFD) based contaminant fate and transport model. To ensure online tractability, the framework uses deep Gaussian process (DGP) based emulator of the multizone-CFD model. To effectively represent the transient response of the multizone-CFD model, the DGP emulator is reformulated using a matrix-variate Gaussian process prior. The resultant deep matrix-variate Gaussian process emulator (DMGPE) is used to define the likelihood of the Bayesian framework, while Markov Chain Monte Carlo approach is used to sample from the posterior distribution. The proposed method is evaluated for single and multiple contaminant sources localization tasks modeled by CONTAM simulator in a single-story building of 30 zones, demonstrating that proposed approach accurately perform inference on locations of contaminant sources. Moreover, the DMGP emulator outperforms both GP and DGP emulator with fewer number of hyperparameters.

Automated Contaminant Source Localization in Spatio-Temporal Fields: A Response Surface and Experimental Design Approach

We propose a contaminant detection methodology suitable for robotic automation, which is able to not only locate the source(s) of the contaminant but also estimate its intensity in an environment that is allowed to evolve over both space and time. The essential idea is to flexibly model the contaminant field surface nonlinearly via radial basis functions and to utilize basic notions from the statistical design of experiments concerning optimal placement of observations in order to make incremental decisions about robot movements. Algorithms are presented for determining such movements and the subsequent collection of measurements in three different cases corresponding to different modes of spatio-temporal evolution. The result is an iterative scheme that gradually locates the peaks (sources), as well as the entire contaminant surface. The performance of the method is assessed through simulations from known surfaces. Theoretical issues concerning convergence of parameter estimates in a multiple robots scenario are examined. The method can accommodate measurement noise and does not rely on surface gradient information.

Bayesian-Based Methods for the Estimation of the Unknown Model’s Parameters in the Case of the Localization of the Atmospheric Contamination Source

AbstractIn many areas of application it is important to estimate unknown model parameters in order to model precisely the underlying dynamics of a physical system. In this context the Bayesian approach is a powerful tool to combine observed data along with prior knowledge to gain a current (probabilistic) understanding of unknown model parameters. We have applied the methodology combining Bayesian inference with Markov chain Monte Carlo (MCMC) to the problem of the atmospheric contaminant source localization. The algorithm input data are the on-line arriving information about concentration of given substance registered by distributed sensor network. We have examined different version of the MCMC algorithms in effectiveness to estimate the probabilistic distributions of atmospheric release parameters. The results indicate the probability of a source to occur at a particular location with a particular release rate.