基于变量选择的锅炉NO_x排放的最小二乘支持向量机建模

Least Squares Support Vector Machine Modeling on NO_x Emission of Boilers Based on Variable Selection

电站燃煤锅炉是大气NOx污染的主要来源之一,建立有效的NOx排放模型是锅炉优化降低NOx的基础。针对热工过程变量之间的强相关和耦合性,利用偏最小二乘方法(partial least squares,PLS)对多工况实炉热态测试数据进行重要变量(variable importance in projection,VIP)信息提取和变量选择(variable selection,VS),把最优的变量子集作为最小二乘支持向量机(least squares support vector machine,LSSVM)的输入,最终得到NOx排放的VS-LSSVM模型。最优的输入变量个数通过留一交叉验证法获取。并将该模型与其他建模方法进行对比,结果表明通过变量选择后建模可以降低模型的复杂度,提高模型的泛化能力。

Coal-fired boiler is the main source of NOx pollution emission.An effective model of NOx emission is the basis to reduce NOx emission.Considering the strong correlations and coupling of input variables in thermal power process,partial least squares（PLS） method was applied to extract information of variable importance in projection（VIP） and select variables based on the real test data of different operation conditions,and then,the optimal variables set was taken as input of least squares support vector machine（LSSVM） algorithm.Finally,the variable selection-least squares support vector machine（VS-LSSVM） model between boiler operation variables and NOx emission was established.In this process,optimal number of input variables was obtained by the leave one out cross validation method.The comparison result to other modeling methods indicates that using the method proposed,model complexity is reduced and generalization capacity of model is enhanced.