[关键词]
[摘要]
为了准确测量锅炉出口的NOx排放浓度,针对燃煤锅炉的复杂非线性,提出了一种基于非线性高斯混合回归(Nonlinear Gaussian Mixture Regression, NGMR)的NOx排放浓度预测方法。采用滑动时间窗方法,结合奇异值分解实现稳态判定;进一步采用互信息(Mutual Information, MI)判断不同变量与NOx排放浓度的相关性,确定模型输入变量;利用选定的输入变量,基于NGMR建立NOx排放浓度预测模型;基于某660 MW燃煤机组运行数据,将提出的NGMR模型分别与人工神经网络(Artificial Neural Network, ANN)模型、支持向量回归(Support Vector Machine, SVR)模型以及极限学习机(Extreme Learning Machine, ELM)模型进行对比分析。结果表明:NGMR模型预测均方根误差(Root Mean Square Error, RMSE)为4.66 mg/m3,平均绝对误差(Mean Absolute Error, MAE)为398 mg/m3;绝对误差系数(R2)为0.9;十折交叉验证结果也表明NGMR模型具有良好的预测精度和泛化能力。
[Key word]
[Abstract]
In order to accurately measure the concentration of NOx at the boiler outlet, aiming at complex nonlinearity of coalfired boiler, a NOx concentration prediction model based on nonlinear Gaussian mixture regression (NGMR) was proposed. A time window sliding method, along with singular value decomposition, was used to realize steady judgement; furtherly, mutual information (MI) was adopted to identify the correlation between different variables and NOx emission concentrations, thereby the input variables of model were determined; with the selected input variables, a NOx concentration prediction model was established based on NGMR; based on the operation data of a 660 MW coalfired power unit, the proposed NGMR model was compared and analyzed with artificial neural network (ANN), support vector machine (SVR) and extreme learning machine (ELM) models. The results show that for NGMR, root mean square error (RMSE) is 4.66 mg/m3, average absolute error (MAE) is 3.98 mg/Nm3, and absolute error coefficient (R2) is 0.9. And the 10fold crossvalidation results also show that NGMR model has better prediction accuracy and generalization ability.
[中图分类号]
TK221
[基金项目]