APP下载

基于 Elastic Net 特征变量选择的 SCR 入口 NOx 软测量模型

2021-09-13王印松陈瑞杰

中国测试 2021年12期

王印松 陈瑞杰

摘要:在传统选择性催化还原反应器( selective catalytic reduction,SCR)入口 NOx 软测量研究中,选取相关变量大多是基于机理分析方法,具有一定主观性。针对这一问题,提出 Elastic Net 方法结合最小二乘支持向量机(least squares support vector machine,LSSVM )的 SCR 入口 NOx 软测量模型。首先采用 Elastic Net 对潜在相关变量进行变量选择,该方法无需机理分析,避免变量选择的主观性。此外,Elastic Net 克服最小絕对收敛和选择算子(least absolute shrinkage and selection operator,LASSO)变量选择时因数据内部存在共线性和群组效应而影响选择效果的问题。然后利用 LSSVM 具有的训练速度较快、泛化性能优良和非线性逼近能力强等优点,建立 Elastic Net-LSSVM 软测量模型。现场数据仿真结果表明:Elastic Net-LSSVM 与 LSSVM 相比,在预测时均方根误差减小8.45%,使预测更准确,验证软测量模型的有效性,可为烟气脱硝系统的控制优化提供参考。

关键词: NOx 软测量; Elastic Net;特征变量选择; LSSVM

中图分类号: TP274.2文献标志码: A文章编号:1674–5124(2021)12–0079–08

Soft sensor model of SCR entrance NOxbased on Elastic Net feature variable selection

WANG Yinsong,CHEN Ruijie

(Department of Automation, North China Electric Power University, Baoding 071003, China)

Abstract: In the traditional selective catalytic reduction(SCR) entrance NOxsoft sensing research, the selection of relevant variables is mostly based on mechanism analysis method, which has a certain subjectivity. To solve this problem, this paper proposes a SCR entrance NOxsoft sensor model based on Elastic Net method and least squaressupportvectormachine(LSSVM). Firstly,ElasticNetisusedtoselectvariablesofpotential characteristic variables. This method does not need mechanism analysis and avoids the subjectivity of variable selection. In addition, Elastic Net overcomes the problem that the selection effect is affected by collinearity and group effect in the selection of variables of least absolute convergence and selection operator (LASSO). Then, the soft sensing model of Elastic Net-LSSVM is established by using the advantages of LSSVM, such as fast training speed, excellent generalization performance and strong nonlinear approximation ability. Field-data simulation results show that compared with LSSVM, Elastic Net-LSSVM can reduce root mean square error by8.45%, which makes the prediction more accurate, and verifies the effectiveness of the soft sensor model, which provides a reference for the control optimization of flue gas denitrification system.

Keywords: NOxsoft sensing; Elastic Net; feature variable selection; LSSVM

0引言

为了降低火电机组污染物排放,烟气脱硝系统的控制优化一直是火电机组的研究热点之一,其中 NOx 的测量是解决控制优化的难点[1]。燃煤电站脱硝过程主要采用 SCR 方法,SCR 烟气脱硝系统入口 NOx 浓度在线监测需要经过烟气管道抽气与烟气分析仪处理,造成的迟延时间普遍高达1 min[2]。使用软测量技术能有效解决迟延造成的影响。

软测量方法属于间接测量方法,它通过分析变量间数学关系建立目标函数的预测模型[2]。针对上述问题,通过建立高精度的 SCR 入口 NOx 浓度软测量模型,可实现对入口 NOx 浓度的间接测量,对控制氮氧化物排放具有重要意义。……

登录APP查看全文