基于Kriging方法的电站锅炉NOx排放自适应建模  

Adaptive Kriging model for NO_x emissions from utility boilers

作　　者：孔亮[1] 丁艳军[1] 张毅[1] 吴昊[2] 张雪[1] 吴占松[1] 

机构地区：[1]清华大学热能工程系,热科学与动力工程教育部重点实验室,北京100084 [2]清华大学计算机科学与技术系,智能技术及系统国家重点实验室,北京100084

出　　处：《清华大学学报（自然科学版）》2008年第5期848-851,共4页Journal of Tsinghua University(Science and Technology)

摘　　要：对复杂多变的热工对象建模是实现良好控制性能的难点,为此提出运用Kriging估计方法建立对象的自适应模型。该法是非参数回归的建模方法,无需确定模型结构和训练,就能实现对未知函数的无偏最优估计。通过对样本空间的实时调整还实现了一种自适应的Kriging模型。选取电站锅炉NOx排放作为建模对象,运用现场试验数据,比较了自适应Kriging模型和神经网络模型对NOx排放的内插和外推预测性能。5组结果显示,神经网络模型的平均预测误差为11.59%,而Kriging模型仅为3.49%。Accurate models are needed for advanced thermal control systems of utility boilers due to the complexities and time-variations of the boiler. An adaptive model for utility boilers was developed using the Kriging estimation method based on spatial statistics, a non-parametric method in which the unbiased optimal estimate for the unknown functions is obtained without model structure selection or training. This paper also presents an adaptive Kriging model that adjusts the sample space in real-time. The predictive capacities of artificial neural networks （ANN） and the Kriging model were evaluated by comparing the predicted NOx emissions of utility boilers with data from a series of field measurements, with both interpolation and extrapolation. The results from five groups show that the average predictive error of the ANN model of 11.59% is reduced to 3.49% by the Kriging model.

关 键 词：电站锅炉 Kriging估计 氮氧化物 模型自适应 

分 类 号：TK39[动力工程及工程热物理—热能工程]