摘要
针对脱硝入口NO_(x)浓度监测值作为脱硝前馈输入导致的喷氨控制滞后问题,提出了基于炉膛参数的脱硝入口NO_(x)浓度CIFE-FOA-DELM预测方法。采用互信息特征选择方法进行预测模型的特征变量筛选;引入经果蝇寻优算法优化的深度极限学习建立NO_(x)浓度预测模型;并利用某660 MW火电机组历史运行数据进行模型验证,与反向传播、支持向量机、深度极限学习机、FOA-SVM模型的预测结果进行对比。结果表明:CIFE-FOA-DELM预测方法具备更高的预测精度,平均绝对百分比误差SMAPE、均方根误差SRMSE、拟合优度R2分别为0.261%、1.384、0.965。与CEMS监测数据对比,脱硝入口NO_(x)浓度预测值提前了180 s,有利于解决喷氨控制滞后问题。
Aiming at the lag problem of ammonia injection control caused by the monitoring value of denitrification inlet NO_(x)concentration as the feed-forward input of denitrification,the CIFE-FOA-DELM prediction method of denitrification inlet NO_(x)concentration based on furnace parameters is proposed.A mutual information feature selection method is used to select feature variables for the prediction model;deep limit learning optimised by Drosophila optimisation algorithm is introduced to establish the NO_(x)concentration prediction model;and the model is validated by using the historical operation data of a 660 MW thermal power unit,and the prediction results are compared with those of the backpropagation,support vector machine,deep limit learning machine,and FOA-SVM models.The results show that the CIFE-FOA-DELM prediction method has higher prediction accuracy,and the mean absolute percentage error(SMAPE),the root mean square error(SRMSE),and the goodness of fit(R2)are 0.261%,1.384%,and 0.965%,respectively,and the prediction of the denitrification inlet NO_(x)concentration is 180 s ahead of schedule when compared with the CEMS data,which is conducive to solving the ammonia injection control lag problem.The problem of ammonia injection control lag is solved.
作者
董威
林子杰
王雅昀
DONG Wei;LIN Zijie;WANG Yayun(Shanghai Jinyi Testing Technology Co.,Ltd.,Shanghai 200000,China;China Energy Science and Technology Research Institute Co.,Ltd.,Nanjing 210023,China)
出处
《电力科技与环保》
2024年第3期313-320,共8页
Electric Power Technology and Environmental Protection
基金
国家重点研发计划(2022YFC3701504)。
