基于堆叠泛化集成模型的脱硝系统入口处氮氧化物浓度预测研究  

Prediction of NOx concentration at the entrance of denitration system based on stacking generalization ensemble method

作　　者：马立增 张玲 谷宇 吴俣 唐媛媛 MA Lizeng;ZHANG Ling;GU Yu;WU Yu;TANG Yuanyuan(China Energy BengBu Power Generation Co.,Ltd,Bengbu 233411,China;Nanjing Guodian Environmental Protection Technology Co.,Ltd,Nanjing 210061,China)

机构地区：[1]国能蚌埠发电有限公司,安徽蚌埠233411 [2]南京国电环保科技有限公司,江苏南京210061

出　　处：《电力科技与环保》2022年第6期517-524,共8页Electric Power Technology and Environmental Protection

基　　金：国能蚌埠发电有限公司科技项目(GN蚌发采[2022]121号)。

摘　　要：为提高燃煤电厂选择性催化还原法脱硝入口处氮氧化物浓度的预测准确度,考虑锅炉燃烧过程中非线性、工况复杂的特点,使用某电厂660 MW火电机组的分布式控制系统的历史运行数据进行模拟实验,提出了一种基于堆叠泛化集成模型的氮氧化物浓度预测方法。首先,采用随机森林特征选择算法选择出对脱硝入口处氮氧化物浓度有较大影响的辅助特征变量。然后,利用选定的特征变量构建多模型堆叠泛化集成氮氧化物预测模型。其中,以反向传播神经网络、支持向量回归和决策树为基础模型,线性回归为元模型。预测结果表明,与单模型相比,集成模型有更高的预测准确率(89.2%)和决定系数R2(91%),且通过十折交叉验证也表明集成模型的均方根误差均低于三种单一模型的均方根误差,具有较强的鲁棒性和泛化能力。所提出的算法和模型适用于预测脱硝入口处氮氧化物浓度预测。Accurate prediction of NOxconcentration at the inlet of selective catalytic reduction denitration process plays an important role in controlling NOxemission of coal-fired power plants.Therefore,in the view of the nonlinear and complex working conditions in the boiler combustion process,a NOxconcentration prediction method based on stacking generalization ensemble method was proposed in this paper based on the simulation experiment of the historical operation data of the distributed control system of a 660MW thermal power unit.First,the random forest feature selection algorithm is used to select the auxiliary feature variables that have a greater impact on the NOxconcentration at the denitrification inlet.Then,a multi-model NOxprediction model based on stacking generalization is constructed by using the selected variables,in which the back propagation neural network,support vector regression and decision tree are the base-learner,and linear regression is the meta-learner.The prediction results show that,compared with the single model,the integrated model is better with higher prediction accuracy(89.2%)and decision coefficient R~2(91%),and ten-fold cross validation also shows that the root mean square error of the integrated model is lower than the other three,which has a strong robustness and generalization ability.Therefore,the algorithm and model proposed in this paper have a good effect in predicting the concentration of NOxat the entrance of denitrification.

关 键 词：脱硝入口处氮氧化物浓度 堆叠泛化集成 随机森林特征选择 

分 类 号：X511[环境科学与工程—环境工程]