基于贝叶斯推理的突发水污染事件水质预测模型参数估计

参数的精确性是准确构建突发水污染事件水质预测模型的前提与保障。论文首先根据有限差分法和贝叶斯推理构建水质预测模型参数估计模型,然后通过Metropolis-Hasting抽样方法得到较为合理的参数,最后以发生在某段明渠的突发水污染事件为例,讨论了恒定流与非恒定流两种情景下不同观测噪声对参数估计结果的影响,并与采用有限差分-单纯形法得到的结果进行对比。结果表明:有限差分-贝叶斯方法具有较强的适用性和良好的抗噪声能力,采用该方法能获得较高精度的参数值。该研究为突发污染事件预测模型的构建提供一条新途径。

Water Quality Prediction Model Parameters Estimation for Sudden Water Pollution Accidents Based on Bayesian Inference

The accuracy of parameters is the premise and guarantee for accurately constructing the water quality prediction model of sudden water pollution accidents. This paper first constructs the parameter estimation model by Finite Difference Method and Bayesian inference, then obtains reasonable model parameters thorough Metropolis-Hasting sampling method, and finally taking a sudden water pollution accident in a certain open channel as an example, the effects of different observation noises on the calibration results are discussed under the two scenarios of control flow and non-control flow by comparing with the results obtained by the finite difference-simplex method. The results show that the finite difference-Bayesian method has strong applicability and a good anti-noise, which gives a high precision parameter value. It provides a new way for constructing the predication model of sudden pollution accidents.