近年来，随着我国经济快速增长和能源消耗量增大，区域大气污染问题日益严峻。而区域大气污染与污染物排放、区域大气环流和污染物传输有关。针对区域大气污染的形成过程、影响因素和污染物分布的数值模型研究对排放控制措施的制定具有重要意义。同时，我国城镇化的快速发展导致位于郊区的工业园区逐渐被城镇包围，安全防护距离不足的问题凸显，化工气体泄漏事故对周边人口的威胁增大。因此，对气体泄漏事故后果的快速判断有助于制定合理的化工事故应急响应策略。由于气体泄漏事故与区域大气污染的形成过程、数值模拟的理论相似，尺度不同，本文针对不同尺度气体扩散计算的特点，选择合适的气体扩散模型模拟区域大气光化学污染过程和局部气体泄漏事故并评估环境影响。针对以近地面臭氧为代表的区域大气光化学污染物的形成过程，使用以欧拉三维扩散模型为核心的综合空气质量扩展模型（CAMx）对北京地区2011年冬季大气重度污染事件进行模拟。模拟结果显示低风速、强太阳辐射和高浓度的一次污染物是导致北京中心城区高浓度臭氧的直接原因，北京地区臭氧浓度日变化规律符合文献报道。对于开阔地区的有毒/可燃气体泄漏事故，本文提出了一种结合气体传感器、气体扩散模型和人工神经网络的气体泄漏事故环境影响快速评估方法。由于传统的气体扩散模型需要详细的泄漏源随时间变化信息（稳态/瞬态）来计算气体扩散的影响范围和浓度分布，而泄漏源信息往往在事故状态下难以获取，限制了气体扩散模型在事故环境影响快速评估以及应急响应决策制定等领域的应用。使用本文提出的快速评估方法能够在泄漏源信息未知的情况下凭借气体传感器的报警信息和实时气象信息使用经过训练的人工神经网络快速计算环境敏感位点的泄漏气体浓度和气体扩散时间，为事故应急响应决策的制定提供必要的辅助。对于复杂地表的有毒/可燃气体泄漏事故，CFD模拟能够提供与扩散有关的全部信息。由于障碍物的存在对局部风场的随机扰动造成气体浓度分布具有时间上不稳定，空间上不对称的特征，无法使用少数位置的浓度探测值对整个空间的分布进行预测。为此，本文提出了一种基于优化基函数的数据驱动函数拟合方法，对一维浓度-时间动态数据进行拟合，分析基函数类型与拟合数据长度对拟合函数外推性质的影响，并在此基础上探索基于优化基函数的事故后果快速评估方法。
In the recent years, the rapid development of China's economy and industrialization has led to a big increase of energy consuming and emission of air pollutants.Regional air quality problem caused by emissions, regional atmospheric circulation and air pollution dispersion is becoming a serious issue. The numerical modeling of regional air quality, in terms of emissions, atmospheric chemistry process and transportation of air pollutants, plays an important role in developing emission control policies. In the mean time, due to the progress of urbanization in mid-sized cities in China, the protection distances between chemical plants and residential areas are often not sufficient to protect local residents from accidental chemical spills. The numerical modeling of gas dispersion from chemical incident and real-time assessment of its consequence are essential for emergency response and preparedness. The dispersion of released gases and regional air pollutants are similar in basic numerical models and processes but are different in atmospheric scales. In this thesis, appropriate numerical gas dispersionmodels are selected for the simulation of regional air pollution and local scale chemical spills and their environmental impacts are also evaluated.Ground level ozone is one of the typical secondary photochemical pollutants in regional scale. A Euler 3-dimensional comprehensive air quality model with extensions (CAMx) is implemented for the simulation of winter ozone episodes in Beijing domain. The spatial distribution of ozone is found to be strongly determined by regional meteorological status with slow wind profile, high photolysis rate and massive emission of primary air pollutants. The results from daily ozone concentration forecasts have shown acceptable agreement with that in literature reviews.A real-time consequence analysis method that estimates hazardous gas dispersion on a flat terrain is developed by the integration of gas detectors, neural network and gas dispersion models. The conventional (heavy) gas dispersion models used on local and small scales requires the knowledge of time-dependent release velocity to calculate the spatial and temporal distribution of released chemicals. However, in the real case it is very difficult to acquire the time-dependent release velocity, which limits the application of conventional gas dispersion simulations in emergency preparedness. However, given the real-time readings from gas detectors and local meteorological stations, the real-time consequence analysis method developed in this dissertation makes it possible to estimate hazardous gas concentrations at distant off-site locations without the need of knowing the actual release rate while providing technical support for emergency decision making.As to the ground level disperion of hazardous chemicals on a complex terrain, CFD modeling could provide big data of spatial and temporal distributions of wind speed vectors, gas concentrations and all other necessary parameters. Due to the turbulence and stochasticity of air flow in the tail zone of obstructions, the temporal concentration at fixed location is unstable and the spatial distribution of gas concentration shows no symmetry, making it impossible to prediction the spatial and temporal distribution of gas concentration using limited data acquired from several sampling points. Thus, A data-driving function fitting method is proposed based on the optimized set of basis functions. The 1-dimentional concentration-time data obtained from CFD simulation is used to build the data-driving function using linear least square method. The type of basis function and length of data fitted are analyzed in detail to explore their impacts on the extrapolation performance of the data-driving function. Finally, the concept of fast consequence analysis approach based on the data-driving function fitting method is proposed.
