电站煤粉炉NO_x排放特性的最小二乘支持向量机模型

Least Squares Support Vector Machine Modeling on NO_x Emissions for Pulverized Utility Boilers

为获得电站煤粉炉NOx排放特性的在线预测模型,实现低NOx闭环运行控制,以某电厂300MW四角切圆燃烧煤粉炉为研究对象建立了NOx排放特性的最小二乘支持向量机模型。在建模过程中,进行了模型性能对核函数、惩罚因子γ和核函数参数σ2的敏感性分析,并运用遗传算法和交叉证实获得了γ和σ2的最佳值。最后利用不同试验工况下的样本数据检验了模型的预测性能,并将该模型的预测性能与BP神经网络模型相比较,结果说明该模型的建模特点和预测性能能够满足NOx排放的在线预测。

In order to get an on-line predicting model of NOx emissions and achieve closedloop operation of NO, control, a LS-SVM model of a certain 300MW tangentially pulverized utility boiler was developed. The model performance＇s sensitivity to kernel function, cost error γand kernel function parameter 2 was studied and analyzed. And the optima of 7 and 2 were solved by using genetic algorithm and cross-validation. The model prediction performance was validated by using some test samples and compared to BP neural network. The result showed that LS-SVM＇s feature of modelling and prediction performance could satisfy on-line prediction of NO, emissions.