基于河流示踪实验的Bayes污染溯源:算法参数、影响因素及频率法对比

发布时间：2018-05-30 21:34

  本文选题：贝叶斯推理 + 污染源反演　； 参考：《中国环境科学》2017年10期

【摘要】：基于贝叶斯理论,结合浓度时间序列方差假定和Adaptive Metropolis MCMC后验采样,建立了用于突发水污染应急溯源的贝叶斯推理方法.该方法结合经验知识和监测事实对源项参数的分布进行推理,直接对溯源结果的反向不确定性用概率分布形式进行表征.依据河流实地示踪剂实验案例,对Bayesian推理溯源的实际效果、后验参数相关性和影响因素等方面进行了验证和测试,结果表明源项参数和方差的后验概率密度的偏度对方差假定敏感,且得到关键参数推荐值:使用RMSE为目标函数;异方差假定中稳定化因子λ1为0,λ2为0.1~0.5;AM采样建议比例因子sd选择0.1~0.3.并对贝叶斯方法和传统基于优化的频率法在求解思路、计算过程、溯源结果等角度进行了深层次的辨析.本研究相关结果为贝叶斯推理技术在污染溯源的实际应用中提供了较为重要的参考价值.
[Abstract]:Based on Bayesian theory, the Bayesian inference method for traceability of sudden water pollution emergency is established by combining the variance assumption of concentration time series and Adaptive Metropolis MCMC posteriori sampling. In this method, the distribution of source parameters is inferred by combining empirical knowledge and monitoring facts, and the inverse uncertainty of traceability results is directly represented by probability distribution. Based on the experimental cases of river tracer, the actual effect of Bayesian reasoning tracing, the correlation of posterior parameters and the influencing factors are verified and tested. The results show that the bias difference of the posterior probability density of the source term and the variance is sensitive, and the recommended value of the key parameters is as follows: using RMSE as the objective function; in the heteroscedasticity assumption, the stabilization factor 位 _ 1 is 0, 位 _ 2 is 0.1 ~ 0.5am sampling ratio factor SD is 0.1 ~ 0.3. Furthermore, the Bayesian method and the traditional frequency method based on optimization are analyzed in the aspects of thinking, calculation process and traceability. The results of this study provide an important reference for Bayesian reasoning in the practical application of pollution traceability.
【作者单位】： 哈尔滨工业大学环境学院;南方科技大学环境科学与工程学院;
【基金】：国家水体污染控制与治理科技重大专项基金(2012ZX07205-005) 中国博士后科学基金(2014M551249)
【分类号】：X52
，

本文编号：1956905