中文摘要：

为克服传统燃烧优化算法受制于小样本建模的缺点,提出了一种基于大规模数据的NOx排放特性建模方法。应用核心向量机（core vector machine,CVM）对11660组实验数据、共77维运行参数建立了超超临界锅炉的NOx排放特性模型,并对模型参数C和ε进行优化,选定模型参数组（C,ε）为（105,6×10-6）,得到了较短的建模时间和较高的预测精准度。同时将建立的CVM模型与其他常见算法支持向量机（support vector machine,SVM）和SVMLight进行性能对比,结果表明,CVM具有优越的收敛速度和更强的泛化能力,随着建模数据量的增加,CVM模型预测准确度有所提升,在建模时间上表现平稳,相对于其余2种算法具有显著优势。

英文摘要：

In order to overcome the traditional combustion optimization algorithm subjected to the disadvantage of small sample modeling, a method of large-scale data modeling on NOxemission property was proposed. Using a core vector machine and based on 11660 sets of experiment data and 77 dimensions of operating parameters, we established a NOx emission property model of ultra-supercritical boilers, and optimized the model parameter C and ε. When the selected parameter set（C,ε） was（105,6×10-6）, a shorter modeling time and higher prediction accuracy can be achieved. Then we compared the core vector machine with other modeling algorithms（SVM and SVMLight）, and the results show that the core vector machine model has a faster convergence rate and stronger generalization ability. With the increase in modeling data scale, the core vector machine model＇s prediction performance has been improved. Its modeling time exhibits stable properties and significant advantages over the other two algorithms.