基于高斯过程回归的燃煤烟气汞排放预测  

Prediction of Mercury Emission from Coal-fired Flue Gas Based on Gaussian Process Regression

作　　者：钟美[1] 赵兵涛[1] 黄朔[1] 

机构地区：[1]上海理工大学能源与动力工程学院,上海200093

出　　处：《动力工程学报》2016年第12期987-992,共6页Journal of Chinese Society of Power Engineering

基　　金：国家自然科学基金资助项目(50806049;51278095);上海市自然科学基金资助项目(08ZR1415100);沪江基地建设资助项目(D14001)

摘　　要：选取对烟气汞排放影响显著的特征参数,采用一类新的随机过程方法——高斯过程回归模型来预测烟气中单质汞、氧化汞和颗粒汞的排放浓度,分别讨论了协方差函数和样本比例对模拟预测精度的统计学影响.结果表明:平方指数协方差函数优于有理二次协方差函数和Matern协方差函数;预测精度随样本比例的增大而提高;高斯过程回归模型优于常规非线性模化方法并显示出更好的鲁棒性,对烟气中汞的形态预测有较好的适用性.By selecting the parameters which significantly influence the mercury emission from coal-fired flue gas, the concentration of elemental, oxidized and particulate mercury in flue gas was predicted using Gaussian process regression, a new random process method, while the effects of covariance function and train-test sample ratio on the simulation accuracy were respectively studied. Results show that the squared exponential covariance function is better than rational quadratic and Matern covariance function; the predicted accuracy increases with the rise of train-test sample ratio; Gaussian process regression is superior to traditional modeling methods of nonlinear regression, and displays good generalization ability, which therefore has strong applicability in prediction of mercury speciation in coal-fired flue gas.

关 键 词：燃煤烟气 汞形态 高斯过程回归 模化 预测 

分 类 号：TK16[动力工程及工程热物理—热能工程]