基于神经网络在线学习的脱硝系统入口氮氧化物预测  

Prediction of denitrification system inlet nitrogen oxide based on neural network online learning

作　　者：金秀章[1] 张少康[1] JIN Xiuzhang;ZHANG Shaokang(School of Control and Computer Engineering,North China ElectricPower University,Baoding,071003,China)

机构地区：[1]华北电力大学控制与计算机工程学院,河北保定071003

出　　处：《河北大学学报（自然科学版）》2018年第4期423-431,共9页Journal of Hebei University(Natural Science Edition)

基　　金：国家重点专项资助项目(2016YFB0600701)

摘　　要：针对脱硝系统入口氮氧化物静态软测量预测模型不能满足变负荷时需求的问题,建立了一种基于神经网络在线学习的软测量模型.利用粒子群算法对静态神经网络的参数进行寻优,结合预报误差和当前预测误差的大小在线更新网络的权值、阈值和学习速率,可以满足不同负荷下的需求,利用电厂的实际运行数据对模型进行了验证.结果表明:在不同负荷下,建立的神经网络在线学习模型的准确性高,实时性好,泛化能力强,可以很好地对入口氮氧化物进行预测,为脱硝系统入口氮氧化物在线测量和监测提供了一种有效的方法.To solve the problem that Static soft sensing model of denitrification system inlet nitrogen oxides can notsatisfy the requirement of variable load,a soft sensor model based on neural network online learning was proposed.Finding the optimum solution of Static neural network by particle swarm optimization,combining prediction error with current prediction error to update weights,thresholds and learning rates online,which can satisfy the requirement of variable load,the model was verified with power plant actual data of power plant.Results show that neural network online learning model is high accuracy and good realtime,with strong generalization ability,which is able to predict the inlet nitrogen oxides,and provide a effective method for online measuremet and monitoring of inlet nitrogen oxides in NOxreduction control system.

关 键 词：入口氮氧化物 核主元法 粒子群算法 预报误差 神经网络在线学习 

分 类 号：TK314[动力工程及工程热物理—热能工程]