Parallel Emergency Management Oriented Research Of Atmospheric Dispersion Modeling And Source Estimation Methods

The air pollution brought by the industry production has an adverse impact on social stability and public health.Excessive emission of air contaminant will seriously affect the air quality,and the release accident of the hazardous substance can cause a great amount of death.Therefore,the management of air pollution emissions is of great importance.Due to the lack of field monitoring data and the difficulty of recurring air pollution incidents,the conventional management of the air contaminant emission stays at the qualitative level,relying only on the expert experience or subjective opinions of managers.The parallel emergency management mode provides an effective way for air pollution emission management.With the support of the parallel management system,computational experiments can be implemented to obtain sufficient related data,reveal the mechanism of the atmospheric dispersion,and verify the emergency management plan as well.Additionally,this system can also support online prediction and decision-making during the release accident.Taking the parallel emergency management of air contaminant emission in the chemical industrial park as the research background,this thesis analyzes related research and the development trend of atmospheric dispersion modeling and emission source estimation methods.This thesis involves the dynamic data-driven atmospheric dispersion model,the machine learning dispersion model combined with the mechanism model knowledge,and the source estimation method,as well as the construction of a parallel management prototype system.The main work and innovations of the thesis can be summarized as follows:(1)A dynamic data-driven atmospheric dispersion model is established.The atmospheric dispersion mechanism model remains static during the operation.Therefore,the prediction error tends to accumulate when this model is applied in the dynamic environment.In this thesis,the atmospheric dispersion mechanism model is combined with the dynamic observed data: particle filter is used as the data assimilation method to assimilate real-time observation into the Gaussian plume model.With the data assimilation,the model parameters can be calibrated and estimated dynamically,and the model prediction accuracy is improved consequently.Further,the particle filter is combined with the expectation-maximization(EM)algorithm framework to reduce the difficulty of parameter estimation.The proposed data-driven modeling method provides an effective way for the atmospheric dispersion modeling in a dynamic environment.(2)A machine learning dispersion model with mechanism model knowledge is proposed.When applied in the atmospheric dispersion,single mechanism modeling or data modeling method cannot balance the prediction accuracy and computational efficiency.To deal with the problem,parameters in the Gaussian plume model,which is a mechanism dispersion model,are introduced into the feature selection of two machine learning models(i.e.artificial neural network(ANN)and support vector regression(SVR)).The proposed Gaussian-machine learning prediction model improves the accuracy of the machine learning model with limited computational cost increasing.In addition,the fitting ability and generalization of these two machine learning models are compared and analyzed,which provides guidance for model selection in practical applications.(3)An optimization-based source estimation method using machine learning dispersion model is proposed.The conventional optimization-based source estimation method is constrained by the limited accuracy of the forward dispersion model.In this thesis,the machine learning dispersion model is applied in the particle swarm optimization(PSO)as the forward dispersion model.Compared with the PSO algorithm using the Gaussian dispersion model,the source estimation accuracy of the proposed method is improved.Furthermore,the sensitivity analysis of the source estimation method reveals the influence of two factors(i.e.measurement noise and the density of the sensor network)on source estimation performance.(4)A parallel management prototype system for air contaminant emission is constructed.According to the the parallel emergency management mode,a variety of related theories and methods are integrated to establish a parallel management prototype system for air contaminant emission in a particular chemical industrial park.This prototype system has the abilities of data monitoring and early warning,dynamic simulation,source estimation etc.In order to provide support of data monitoring for the prototype system,an UAV-based monitoring system is developed achieve flexible and efficient data collection.Focusing on parallel emergency management of air contaminant emission,this thesis proposes novel modeling methods for the atmospheric dispersion by combining mechanism modeling and data modeling methods.Further,the proposed atmospheric dispersion model is applied to the source estimation,which improves the source estimation results.Based on the existing theories and methods,a parallel management prototype system for air contaminant emission is constructed,with an UAV-based monitoring system developed.Our research helps managers to understand the air contaminant dispersion incident,and aids the emergency decision-making,which is of great social significance.