数据驱动的化工园区大气污染物泄漏源参数反演研究

Source term estimation of gaseous chemical spills is essential for emergency response and consequence assessment. One of the bottlenecks limiting the development of source term estimation is the complexity and stochasticity of atmospheric dispersion and monitoring. The objective of this project is to build a data-driven gas dispersion model that is able to calculate the gaseous dispersion of hazardous chemicals precisely. The data-driven model is used to improve the performance of source term estimation algorithm. Thus, a series of hypothetical gas release scenarios are generated to provide data for the projects. Firstly, the lattice gas model together with the Eulerian and Lagrangian dispersion theory are coordinated to develop a gas dispersion model based on cellular automata. Secondly, for integrating the monitoring data into the cellular dispersion model, the 4-Dimensional variation (4DVAR) and Ensemble Kalman Filter (EnKF) assimilation algorithms are introduced in implementing the monitoring data to regulate the state parameters of the cellular dispersion model and improve the model predictions. Finally, on this basis, the optimization method based source term estimating algorithm is implemented with the cellular dispersion model and the uncertainty of the estimation is assessed. This project will explore the integration of data assimilation and cellular gas dispersion model as well as its corresponding source term estimation functions. The results will be used to support emergency decision-making and response.

化工园区气体泄漏事故溯源对于事故应急处置和预测事故的后果具有重要意义。然而，大气扩散过程的随机性与复杂性与污染物浓度观测的不确定性是泄漏溯源技术发展的瓶颈。项目以虚构的化工园区一系列不同的泄漏情景为研究对象，首先依据晶格气体模型理论及欧拉和拉格朗日气体扩散理论，开展基于元胞自动机的大气扩散模型研究，保证模型能够适应复杂多变的化工园区扩散环境同时具有较高的计算速度；其次，利用虚构的泄漏情景模拟数据替代实际观测数据，借助资料同化理论，研究以四维变分和集合卡尔曼滤波为核心算法的数据驱动元胞大气扩散模型参数修正方法，探索观测数据在改进模型精度，提高预报准确性中的应用；最后，在上述研究的基础上，结合优化算法，建立基于元胞大气扩散模型的泄漏源参数反演模型并分析反演结果的不确定性。该项目将探索资料同化与元胞大气扩散模型在化工园区尺度下的泄漏源参数估计功能，为事故应急处置提供技术支持。

化工园区气体泄漏事故溯源对于事故应急处置和预测事故的后果具有重要意义。然而，大气扩散过程的随机性与复杂性与污染物浓度观测的不确定性是泄漏溯源技术发展的瓶颈。本项目开发一种能够适应复杂地表环境的元胞扩散模型，以得到依据地面障碍物分布和大气非均匀流动场景下的浓度分布，相同条件下计算速度达到传统CFD模型的2.5倍。针对元胞扩散模型对大气流动数据需求，建立了基于风传感器的大气流场重建模型，并应用FDS进行模型验证。研究的创新之处在于将已有的二维元胞扩散模型推广至三维场景，并综合分析了包含模拟区域内泄漏点源的神经网络设计方法，讨论了不同神经网络输入输出策略对情景的适应性。同时对于元胞扩散模型需要高分辨率气象参数的特点，设计了基于主元分析，ELM回归模型和观测数据的风场重建流程，实现了复杂地表环境的二维风场重建，为后续大气扩散模拟和溯源提供了支撑方法。