【摘要】 针对大型工业袋式除尘器破损滤袋智能反演定位的不确定性量化问题，文中基于贝叶斯定理，将反问题的先验信息、理论模型、观测数据转化为反演解的后验概率分布，从而实现将破损滤袋反演定位的不确定性量化为反演结果的不确定性。首先采用数值模拟技术，建立袋式除尘器气固两相流的正向模拟模型，获得反映其输入输出关系的样本数据，然后训练替代模型来代替正向模拟模型，从而提高反演效率，最后通过对整个模型空间进行抽样得到观测信息下破袋位置的后验概率分布。研究结果表明：30组试验，破损滤袋定位的准确率为100%,滤袋破口面积反演结果的平均相对误差为1.50%,最大相对误差为4.15%,最小相对误差为0.02%,抽样数据的方差平均值为17.20,方差最大值为75.15,反演结果误差较小，置信区间范围较窄，抽样数据离散程度较低；将传感器测量误差及单一传感信息造成的不确定性，量化为反演结果的准确率、相对误差、方差及置信区间大小，避免了反演结果的不适定性。与确定性方法相比，贝叶斯反演方法能够实现破损滤袋定位及破口面积判断的同时对反演结果进行不确定性量化，为研究人员提供更加丰富的破袋信息。更多还原

【Abstract】 Since the intelligent inversion positioning of damaged filter bags of large industrial bag filters suffer from uncertainty quantification, in this article, based on the Bayesian theory, the prior information, theoretical model, and observation data of the inverse problem are transformed into the posterior probability distribution of the inversion solution, so as to quantify the uncertainty of inversion positioning of the damaged filter bags as the uncertainty of the inversion result. Firstly, the technology of numerical simulation is adopted to set up the forward simulation model of the gas-solid two-phase flow of the bag filter, in order to obtain the sample data reflecting the input-and-output relationship; then, the surrogate model is used to replace the forward simulation model, thereby improving the inversion efficiency. Finally, by sampling the entire model space, the posterior probability distribution of the damaged bag’s position with the help of the observed information is identified. The results show that in the 30 experiments, accuracy of positioning of the damaged filter bags is 100%; the average relative error of the inversion results of the damaged areas of the filter bags is 1.50%. The maximum relative error is 4.15%, while the minimum relative error is 0.02%. The average variance of the sampled data is 17.20, with the maximum variance of 75.15; the error of the inversion result is small, the confidence interval of the sampled data is narrow, and the degree of dispersion is low. The quantitative analysis on uncertainty caused by the sensor’s measurement error and the single sensor information is described as accuracy, relative error, variance, and confidence interval of the inversion result, which avoids ill-posedness. Compared with the deterministic method, the Bayesian inversion method realizes positioning of damaged filter bags and judgment of damaged areas on one hand and quantifies uncertainty of inversion results on the other hand. It provides researchers with more information about damaged filter bags.更多还原