基于长短时记忆神经网络的锅炉多参数协同预测模型  

Multi-parameter collaborative prediction model of boilers based on long-short-term memory neural network

作　　者：金志远 李胜男 谭鹏[1] 严杏初 刘宇浓 张成[1] 陈刚[1] JIN Zhiyuan;LI Shengnan;TAN Peng;YAN Xingchu;LIU Yunong;ZHANG Cheng;CHEN Gang(State Key Laboratory of Coal Combustion,Huazhong University of Science and Technology,Wuhan 430074,China;Shajiao C Power Plant of Guangdong Energy Group Co.,Ltd.,Dongguan 523900,China)

机构地区：[1]华中科技大学能源与动力工程学院煤燃烧国家重点实验室,湖北武汉430074 [2]广东省能源集团有限公司沙角C电厂,广东东莞523900

出　　处：《热力发电》2021年第5期120-126,共7页Thermal Power Generation

基　　金：中央高校基本科研业务费资助(HUST:2020kfyXJJS030);中国博士后科学基金(2018M632852)。

摘　　要：锅炉协同控制是提高其灵活运行下蒸汽温度平稳的有效手段。以某660 MW燃煤锅炉为研究对象,利用其历史运行数据,建立基于长短时记忆(LSTM)神经网络的主蒸汽温度、再热蒸汽温度、炉膛出口NO_(x)质量浓度、炉膛出口CO质量浓度协同预测模型。模型预测结果表明,该协同预测模型4个输出的相关系数均大于0.94,模型综合预测效果良好,且有较好的泛化能力。该模型为锅炉蒸汽温度、NO_(x)、炉效协同优化控制提供了依据。Cooperative control of boiler is an effective approach to improve steam temperature stability under flexible operation.Taking a 660 MW coal-fired boiler as the research object,by using the historical operation data,a cooperative prediction model based on the long-short-term memory(LSTM)neutral network for main steam temperature,reheated steam temperature,NO_(x) and CO mass concentration at the furnace outlet was established.The prediction results show that,all the correlation coefficients of the four outputs of the cooperative prediction model are larger than 0.94,indicating the model has good comprehensive prediction effect and generalization ability.The research provides a foundation for collaborative optimization control of boiler steam temperature,pollutants emission reduction and furnace efficiency enhancement.

关 键 词：燃煤锅炉 LSTM神经网络 蒸汽温度 NO_(x)/CO质量浓度 多参数协同 预测模型 

分 类 号：TK39[动力工程及工程热物理—热能工程]