摘要
为了实现机组超低排放并能够安全稳定运行,基于脱硝系统入口烟气温度、入口O2体积浓度、入口和出口NOx质量浓度、喷氨量和空气预热器差压等分散控制系统(DCS)数据,分别构建了用于预测氨逃逸的长短期记忆网络(LSTM)模型和支持向量回归(SVR)模型。根据现场测试的SCR系统入口和出口NOx质量浓度和氨逃逸量,计算得到宏观的脱硝装置潜能,结合同时间段内的DCS数据计算氨逃逸量,并作为真实值与LSTM模型和SVR模型的预测值进行对比。结果表明:SVR模型对氨逃逸的预测有较高的准确度和泛化能力,SVR模型对测试样本的均方根误差δMRE=0.007 1μL/L,平均绝对误差δMAE=0.002 4μL/L;LSTM模型对测试样本的预测误差δMRE=0.047 0μL/L,δMAE=0.019 0μL/L。
To achieve safe and stable operation of thermal power plant with ultra-low emissions,a long short-term memory(LSTM)model and a support vector regression(SVR)model for predicting ammonia escape were constructed respectively,based on the distributed control system(DCS)data including the flue gas temperature and O2 concentration at the denitration inlet,the NO,concentration at the denitration inlet and outlet,the amount of ammonia injection,and the pressure difference of the air preheater.According to the NOr concentration and ammonia escape of the SCR denitration inlet and outlet tested on site,the macroscopic denitration device potential was calculated,the ammonia escape concentration was calculated in combination with the DCS data in the same period,which was used as the real value compared with the forecasting results of the LSTM model and the SVR model.Results show that the SVR model has high accuracy and generalization ability for the prediction of ammonia escape.The prediction errors of the SVR model for the test samples are:OmRE=0.0071μL/L,OMAE=0.0024μL/L.The prediction error of the LSTM model for the test sample are:OMRE=0.0470μL/L,omAE=0.0190μL/L.
作者
谭增强
牛拥军
李元昊
曲飞雨
TAN Zengqiang;NIU Yongjun;LI Yuanhao;QU Feiyu(Xi'an West Heat Boiler Environmental Engineering Co.,Ltd.,Xi'an 710054,China;Cangzhou China Resources Thermal Power Co.,Ltd.,Cangzhou 0610o0,Hebei Province,China)
出处
《动力工程学报》
CAS
CSCD
北大核心
2023年第7期917-922,共6页
Journal of Chinese Society of Power Engineering
基金
国家自然科学基金资助项目(51976072)
华能集团总部科技资助项目“基础能源科技研究专项”(HNKJ20-H50)。
关键词
火电厂
氨逃逸
预测模型
LSTM
SVR
thermal power plant
ammonia escape
predictive model
LSTM
SVR