摘要
针对燃煤电厂SCR脱硝反应器出口氮氧化物浓度测量存在较大迟延,以及吹扫阶段数据失真等问题,提出一种基于改进哈里斯鹰算法(Improved Harris hawk optimization,IHHO)和长短时记忆神经网络(Long short-term memory neural network,LSTM)的NOx浓度预测模型。首先利用最大信息系数对不同输入变量进行迟延时间估计,并通过最大相关最小冗余算法完成输入变量的选择并重构输入特征序列。然后,使用重构数据来完成LSTM模型的建立,并利用IHHO对网络相关参数进行优化。最后将所建模型的预测结果与未进行特征选择的IHHO-LSTM模型、进行特征选择后的HHO-LSTM模型和LSTM模型进行验证对比。结果表明,相较于其他预测模型,所提的IHHO-LSTM模型预测精度更高,具有较好的动态特性。
In view of the problems of the large delay in measuring the nitrogen oxide concentration at the exit of SCR denitrification reactors in coal-fired power plants,as well as the distortion of the data during the purge phase,a NOx concentration prediction model based on improved Harris hawk optimization(IHHO)and long short-term memory neural network(LSTM)is proposed.At first,the delay time of different input variables is estimated by the maximum information coefficient,and the input variables are selected and input feature sequence is reconstructed by the maximum correlation and minimum redundancy algorithm.Then,the LSTM model is established by using the reconstructed data,and the network parameters are optimized by IHHO.Finally,the predicted results of the constructed model are compared with IHHO-LSTM model without feature selection,HHO-LSTM model after feature selection and LSTM model.The results show that the IHHO-LSTM model has higher prediction accuracy and better dynamic characteristics than other prediction models.
作者
许子明
姜浩
赵文杰
XU Ziming;JIANG Hao;ZHAO Wenjie(Department of Automation,North China Electric Power University,Baoding 071003,China)
出处
《电力科学与工程》
2023年第8期71-78,共8页
Electric Power Science and Engineering
关键词
SCR脱硝系统
迟延估计
变量选择
长短时记忆神经网络
哈里斯鹰优化
NOx浓度预估
SCR denitrification system
delay estimation
variable selection
long short-term memory neural network
Harris hawk optimization
estimation of NOx concentration