Research On Source Identification Approaches For Hazardous Chemical Releases

Accurate and timely evaluation of the strength and location of pollutantsources plays an important role in emergency responses involving hazardouschemicals, particularly when toxic gases are released. An effective way is touse concentration observations, in and around the accident scenes, combinedwith a dispersion model to construct an inversion model.Basing on observations of the concentration and atmospheric environmentcondition, dispersion modeling and inversion methods, the research wasconducted with following steps:(1) Source identification modeling;(2) Modelmodification using practical dispersion model;(3) Validation with syntheticdata and practical data. Therefore, the main tasks and findings are as follows:Firstly, source inversion model is constructed by an optimizationframework. The pattern search method is applied to source identification forthe first time. Using the observations of concentration and atmosphericenvironment condition, the source identification problem was transferred intoan optimization problem. Then the pattern search method was employed toadjust parameters to find the optimal matching of calculated and observed concentrations. The idea of neighborhood search of the pattern search methodmade it convenient to combine with other global methods for a more accuratesolution. Then the structure of hybrid optimization method and the timing forthe combination were studied. As a result, a inlaid structure was employed,and the accuracy of the inversion results was improved significantly as well.Secondly, an approach based on Bayesian inference and optimizationmethod is proposed to identify the release source. The priori information ofthe model parameters, as well as the final inversion results are presented byprobability distribution. Based on the Bayesian inference, a posteriordistribution is obtained. The Markov Chain Monte Carlo (MCMC) samplingis employed to sample the posterior distribution so as to get the estimatedvalue of the parameters and to evaluate the error. In order to improve theefficiency of the MCMC sampling, the optimization method is combined withthe Bayesian inference. The optimization method is employed in theinitialization process before the sampling to get a comparatively bettersolution for the MCMC sampling. With the combination of the Bayesianinference and optimization method, it can not only maintain the performanceof the Bayesian inference in solving the uncertainty problem, but also canimprove the computational efficiency.Thirdly, a dispersion model based on cellular automata(CA) is constructed.The model can effectively predict the concentration in the environment at anytime. According to the optimization modeling method and/or Bayesian inference, cellular automata model and the actual observations of theconcentration are used to get a better solution. The validation results show thatthis approach can improve the accuracy of the results, and can reduce theprobability of the error identification.Finally, simulation and field testing was combined in the research forsource inversion validation. The source inversion methods were first validatedwith synthetic data, and then the practical data from field testing were used toverify the validation of the inversion method. A field testing platform isestablished. The concentrations were obtained effectively though thecombination of fixed and mobile monitoring network. The results show thatthe optimization method based approach may cause large error in complexenvironment when the dispersion model is selected improperly. As theobservation error and model error are considered in Bayesian inference, it isable to obtain comparatively better results.Innovation is mainly reflected as follows:First of all, the inversion method is introduced to the source identification,using the concentration observations, the dispersion model, as well as theoptimization algorithms and/or Bayesian inference. The pattern search methodwas used in the source identification for the first time. Based on a frameworkof optimization, different models were established to study the sourceidentification with the concentration observations from different observationnetwork. Secondly, the Bayesian inference is combined with optimization algorithms. The optimization algorithms are used in the initialization processto obtain a better sample for MCMC sampling. The combination of themethods can ensure the performance of the Bayesian inference in solving theuncertainty problem, and the calculation accuracy and efficiency are improvedas well. Thirdly, we introduce the cellular automata to the gas dispersionmodeling. The CA-based dispersion model can predict the variation of theconcentration in different place and time; therefore, the injury scope can beestimated and thus the accident can be effectively controlled.