Study On Designing And Implementing High Performance Atmospheric Dispsersion Model And Source Back-Calculation Model For Hazardous Chemical Release

As chemical industry developed rapidly in recent ten years in China, the number of accidents in chemical industy increased very quickly. Chemical products producing, traspoting and storing process involves large kinds of hazardous material. Thus it will cause serious damage and endanger large number of people. However, because the simulation of atmospheric dispersion procedure requires complex models and huge amount of computations, traditional computation methods are not able to give consideration to both speed and accerucy. So they are not suitable in emergency responsing processes.In emergency reponsing processes, the source parameters, which are essential for consequence predicting models, are difficult to be acquired due to the limitations of field conditions. Thus, investigating source back calculating methods based on filed monitoring are very useful.Based on these considerations, this thesis represents a high performance consequence predicting method and a source determination method.In consequence predicting part, this thesis re-designed a Gaussian multi-puff model. By using technologies of smoke tracing and concentration field referencing, the amount of computation significantly decreased. Then thesis introduces the methods of parallelizing multi-puff model by employing GPU computation and CUDA computation framework. Case studies carried out, proved outstanding acceleration of this model.Based on these researches, this thesis represents the second part, a Markov Chain Mentor Carlo based source parameters back-calculation method. In MCMC process, this research used Gibbs Sampler as the chain constructor. This research successfully calculates source position, length leaking duration and its leaking rate variation. By implementing these methods using Python and C programming language, this thesis introduced GAPS (GPU Accelerated Puff Simulator) and its source inversion module GAPS-Rev. Several cases studies are carried out to investigate its effects and stability. Also, different schemes of input parameters are designed to compare their effects to back-calculating process.