摘要
针对燃煤电厂中选择性催化还原(Selective Catalytic Reduction,SCR)脱硝系统入口NO_(x)浓度的测量传感器迟延大,不能准确反映其浓度的实时变化的问题,提出了利用Copula熵(Copula entropy,CE)筛选与入口NO_(x)浓度软测量相关的辅助变量,利用变模态分解(Variational Mode Decomposition,VMD),将入口NO_(x)浓度分解为不同中心频率的子序列信号,建模充分拟合目标变量的数据特征。采用二级建模方法,第一级,将分解后得到的入口NO_(x)浓度子序列信号分别利用贝叶斯回归算法(Bayesian Regression,Bayes)进行训练并预测,叠加得到完整的预测结果,第二级,对训练中产生的验证集误差值利用Lasso算法建立误差预测模型,得到测试集预测误差的预测值,并与第一级模型得到完整预测结果叠加,实现误差补偿,提升模型预测精度。其中,Bayes及Lasso网络超参数利用天牛群算法进行自动寻优;仿真结果显示,VMD分解并带误差补偿模型对比未经VMD分解带误差补偿模型,Bayes及Lasso单一模型的均方根误差、平均绝对误差、平均绝对百分比误差最小,能够实现对入口NO_(x)浓度的准确软测量。
Addressing the issue of the measurement sensor for inlet NO_(x) concentration in the Selective Catalytic Reduction(SCR)denitration system of coal-fired power plants exhibiting a large delay and inability to accurately reflect the real-time concentration changes,we propose a method utilizing Copula entropy(CE)to screen the auxiliary variables related to the soft measurement of the inlet NO_(x) concentration.Additionally,Variational Mode Decomposition(VMD)is employed to decompose the inlet NO_(x) concentration into subsequence signals with different center frequencies,enabling the modeling of data characteristics that fully fit the target variable.We adopt a two-stage modeling approach.In the first stage,the decomposed inlet NO_(x) concentration subsequence signals are individually trained and predicted using Bayesian Regression(Bayes)algorithm,with the complete prediction results being superimposed to form complete prediction results.In the second stage,we establish an error prediction model for the validation set error generated during training using the Lasso algorithm.The predicted value of the test set prediction error is obtained and superimposed with the complete prediction result from the first-stage model to achieve error compensation,thereby enhancing model prediction accuracy.Among them,Bayes and Lasso network hyperparameters are automatically optimized using the beetle swarm optimization algorithm.The simulation results show that,compared to models without VMD decomposition but with error compensation,as well as Bayes and Lasso single models,the VMD decomposition with error compensation model exhibits the smallest root mean square error(RMSE),mean absolute error(MAE),and mean absolute percentage error(MAPE).This can achieve accurate soft measurement of the inlet NO_(x) concentration.
作者
金秀章
乔鹏
史德金
JIN Xiuzhang;QIAO Peng;SHI Dejin(School of Control and Computer Engineering,North China Electric Power University,Baoding 071003,China)
出处
《华北电力大学学报(自然科学版)》
北大核心
2025年第3期117-124,142,共9页
Journal of North China Electric Power University:Natural Science Edition
