基于特征优化和ISSA-LSTM的脱硝系统入口NOx浓度预测模型
网络安全与数据治理 4期
王渊博,金秀章
(华北电力大学控制与计算机工程学院,河北保定071003)
摘要:针对电厂脱硝系统入口NOx浓度受较多因素的影响波动较大,且CEMS检测仪表有很大迟延难以精准测量的问题,提出了一种基于随机森林算法(RF)和改进麻雀搜索算法(ISSA) 优化长短时记忆神经网络(LSTM)的脱硝系统入口NOx浓度预测模型。首先,通过机理和相关性分析确定与SCR入口NOx质量浓度相关的初始辅助变量,并利用RF算法对辅助变量进行特征优化选择,然后通过互信息(MI)对各辅助变量与输出变量之间进行迟延估计并提取时序特征,并通过小波滤波对输入变量进行降噪处理,建立LSTM神经网络预测模型。利用ISSA算法确定LSTM模型的最优组合参数,最后与传统的LSSVM、RBF、BP模型的预测结果进行对比。实验结果证明,特征优化后的ISSALSTM神经网络预测模型的决定系数(R2)最大,均方根误差(RMSE)和平均绝对百分比误差(MAPE)最小,具备很强的拟合和泛化能力,可以精准预测脱硝系统入口氮氧化物的质量浓度。
中图分类号:TP183
文献标识码:A
DOI:10.19358/j.issn.2097-1788.2023.04.012
引用格式:王渊博,金秀章.基于特征优化和ISSALSTM的脱硝系统入口NOx浓度预测模型[J].网络安全与数据治理,2023,42(4):70-77,84.
文献标识码:A
DOI:10.19358/j.issn.2097-1788.2023.04.012
引用格式:王渊博,金秀章.基于特征优化和ISSALSTM的脱硝系统入口NOx浓度预测模型[J].网络安全与数据治理,2023,42(4):70-77,84.
Prediction model of NOx concentration at the inlet of the denitration system based on feature optimization and ISSALSTM
Wang Yuanbo,Jin Xiuzhang
(School of Control and Computer Engineering, North China Electric Power University, Baoding 071003, China)
Abstract:Aiming at the problem that the NOx concentration at the inlet of the denitrification system in power plants is greatly affected by many factors and fluctuates greatly, and the CEMS detection instruments have great delays and are difficult to accurately measure, a prediction model for the NOx concentration at the inlet of the denitrification system based on the random deep forest algorithm (RF) and the improved sparrow search algorithm (ISSA) optimized longterm and shortterm memory neural network (LSTM) was proposed. Firstly, the initial auxiliary variables related to the mass concentration of NOx at the SCR inlet were determined by mechanism and correlation analysis, and the auxiliary variables were selected for feature optimization using the RF algorithm, then the delay between each auxiliary variable and the output variables were estimated by mutual information (MI) and the timing features were extracted, and the LSTM neural network prediction model was established by denoising the input variables through wavelet filtering. The modified sparrow search algorithm was used to determine the optimal combination parameters of the LSTM model and finally contrasted with the prediction results of the traditional LSSVM, RBF and BP models. The experimental results proved that the ISSALSTM neural network prediction model after feature optimization had the largest coefficient of determination (R2) and the smallest root mean square error (RMSE) and mean absolute percentage error (MAPE), which exhibited strong fitting and generalization ability to accurately predict the mass concentration of NOx at the inlet of the denitrification system.
Key words :NOx concentration prediction;feature optimization; mutual information; sparrow search algorithm;LSTM neural network;random forest
0 引言
为了实现碳中和的目标,我国近年来积极推进能源转型,优化能源结构。根据国家统计局最新公布的数据,2022年火电的装机容量仍然占比52%左右,是我国发电领域中的领头羊。火力发电机组的主要燃料来源是煤炭,而煤炭在燃烧过程中会产生大量的NOx,NOx是造成大气污染的主要污染物之一。
当前我国电厂常用的烟气脱硝方法主要分为两种,分别为选择性催化还原(SCR)脱硝系统和选择性非催化还原(SNCR)脱硝系统。两种方法各有优劣, 前者具有工艺成熟、安全稳定且脱硝效率超过90%等优点,是当前电厂烟气脱硝技术的首选,后者由于脱硝效率低,在烟气脱硝中一般只用作辅助手段。本文研究的燃煤电站采用SCR技术对尾部烟气中的氮氧化物进行脱销处理。
由于燃煤电站锅炉燃烧系统是一个具有大延迟、大惯性的非线性系统,SCR入口NOx浓度容易受不同因素的影响而波动较大,使得精准SCR入口氮氧化物浓度的获取变得困难,进而很难对喷氨量进行精准的控制。喷氨量过低,脱销效果不好,会造成NOx排放不达标;过量喷氨不但影响脱硝效率,又造成巨大的资源消耗,提高运行成本。因此,建立精准有效的脱硝系统SCR入口氮氧化物预测模型,不仅可以帮助脱硝系统精准调控喷氨量,提升脱硝品质,又可以降低电厂的脱硝成本。
本文详细内容请下载:https://www.chinaaet.com/resource/share/2000005271
作者信息:
王渊博,金秀章
(华北电力大学控制与计算机工程学院,河北保定071003)
此内容为AET网站原创,未经授权禁止转载。