基于后向传播神经网络的循环流化床锅炉SO_(2)排放预测分析  

Prediction analysis of SO_(2) emissions from circulating fluidized bed boiler based on BP neural network

作　　者：倪绍佑 姚晨蓬 陈玲红[1] 吴学成[1] 岑可法[1] NI Shao-you;YAO Chen-peng;CHEN Ling-hong;WU Xue-cheng;CEN Ke-fa(State Key Laboratory of Clean Energy Utilization,Zhejiang University,Hangzhou 310027,China;Hangzhou Pro-Energy Heat and Power Co.Ltd.Hangzhou 310018,China)

机构地区：[1]浙江大学能源清洁利用国家重点实验室,浙江杭州310027 [2]杭州杭联热电有限公司,浙江杭州310018

出　　处：《能源工程》2021年第6期31-37,共7页Energy Engineering

基　　金：国家自然科学基金资助项目(51876190)。

摘　　要：基于燃煤热电厂分散控制系统记录的循环流化床锅炉运行状态参数,利用主成分分析对运行数据进行分类和相关性验证。选取热电厂秋季运行的一次风量、给煤量等表征燃烧运行状态的4000组数据建立训练和验证数据库。采用后向传播神经网络建立锅炉出口处污染物SO;排放量的预测模型,研究该处SO;浓度与炉内外运行参数间的非线性关系和模型的局部最优问题,并对比分析不同激活函数、训练次数的预测效果。结果表明:以给煤量、蒸发量和运行温度等参数作为输入时,选用单隐含层、relu函数、学习率为0.005的模型能够较好地预测锅炉出口SO;排放情况,相对误差在5%以下,可为实际运行中燃烧优化控制提供科学依据。The operating data from the distributed control system in a CFB boiler were classified and conducted correlation verification by principal component analysis.4000 sets of data with respect to primary air volume,coal feed volume were selected to establish a training and testing database,representing the combustion operation status in autumn.The backward propagation neural network analyzing was used to establish a prediction model for SO_(2) emissions at outlet of the boiler,to establish the nonlinear relationship between the SO_(2) concentration and the operating parameters inside and outside the furnace.The prediction results of different activation functions and training times were compared and analyzed.The results showed that the relative error is below 5%when the coal supply,boiler evaporation and boiler temperature were used as inputs,and single hidden layer,relu function and learning rate of 0.005 were applied to predict SO_(2) emissions,which provided a reference for practical combustion and operation optimization control.

关 键 词：分散控制系统 主成分分析 BP神经网络 SO_(2)排放预测 

分 类 号：X701.3[环境科学与工程—环境工程]