基于深度神经网络的脱硫系统预测模型及应用

doi:10.16085/j.issn.1000-6613.2020-0751

摘要/Abstract

摘要：

建立了一个隐含层包含一个长短期记忆层（long-short term memory, LSTM）、两个线性整流函数层（rectified linear unit, ReLU）、两个全连接层（fully connected layer）和输入、输出层组成的深度神经网络，用于脱硫系统主要指标预测。该模型对输入参数采用了指数滑动平均、合并最小分析周期等数据预处理技术进行降噪，在网络训练过程中采用dropout技术防止过拟合。仿真结果对比现场数据表明，模型对浆液pH、出口SO₂浓度和脱硫率均体现出良好的预测能力。本文还结合某2×350MW燃煤电厂提供的实际工况数据，以石灰石供浆密度对系统脱硫性能的影响为例，详细介绍了利用所建立的深度神经网络模型测试湿法脱硫系统各参数指标对脱硫效果的影响，并结合化学机理和工业实际进行的诊断过程。

关键词: 燃煤电厂, 脱硫系统, 计算机模拟, 深度学习, 神经网络, 预测, 模型应用, 智慧环保

Abstract:

In this paper, a deep neural network consisting of a LSTM layer, two rectified linear unit layers, and two fully connected layers was established. Data pre-processing such as moving average and minimum analysis period for input parameters to reduce noise was used. During the network training, the dropout technique to prevent overfitting was used. Simulation results and comparison with the existing technology showed that the model has good prediction ability for slurry pH, outlet SO₂ concentration and desulfurization rate. The actual working condition data of a 2 × 350MW thermal power plant and this deep neural network model to test the effect of limestone slurry supply density on the system’s desulfurization performance was also combined.

Key words: coal-fired power plant, desulfurization system, computer simulation, deep learning, neural network, prediction, model application, smart environmental protection

中图分类号:

TP183

马双忱, 林宸雨, 周权, 吴忠胜, 刘琦, 陈文通, 樊帅军, 要亚坤, 马采妮. 基于深度神经网络的脱硫系统预测模型及应用[J]. 化工进展, 2021, 40(3): 1689-1698.

MA Shuangchen, LIN Chenyu, ZHOU Quan, WU Zhongsheng, LIU Qi, CHEN Wentong, FAN Shuaijun, YAO Yakun, MA Caini. Prediction model of FGD system based on deep neural network and its application[J]. Chemical Industry and Engineering Progress, 2021, 40(3): 1689-1698.

图/表 17

表1 神经网络的输入指标和其波动范围

分析指标	波动范围	分析指标	波动范围
浆液pH	4.5~8.5	入口SO₂浓度/mg·m^-3	800~1700
入口烟气流量/m³·h^-1	6×10⁵~1.2×10⁶	出口SO₂浓度/mg·m^-3	0~50
入口烟气粉尘含量/mg·m^-3	30~70	入口烟气氧含量/%	4.94~7.03
石灰石供浆密度/kg·m^-3	1180~1300	浆液密度/kg·m^-3	1100~1150
浆液液位/m	9~10	循环泵开启数量/台	1~4
石灰石供浆流量/m³·h^-1	0~25	石膏排出泵开启数量/台	0~2
入口烟温/℃	90~110	氧化风机流量(湿)/m³·h^-1	5500~5700

图1 数据EMA降噪结果展示

图2 最小分析周期变化对模型性能的影响

图3 模型数据预处理过程

图4 深度神经网络的设计结构

图5 LSTM网络结构的运算流程

图6 Hinton展示的dropout示意图[27]

图7 出口SO2浓度降噪后实际值和预测值的对比

图8 出口脱硫率降噪后实际值和预测值的对比分析

图9 出口SO2浓度降噪后实际值和预测值的对比（局部放大）

图10 浆液pH实际值和预测值的对比分析

表2 神经网络输出指标和其波动范围

预测指标	均方根误差	平均百分误差/%	误差方差
pH	0.0955	0.9726	0.0091
出口SO₂浓度	1.0778	3.6599	1.1551
脱硫率	0.0903	0.0630	0.0081

图11 预测pH和脱硫率误差的频次分布直方图

图12 不同网络预测pH和脱硫率误差主要指标对比

图13 吸收塔石灰石浆液密度对出口脱硫率和出口SO2浓度的影响

图14 监测时段吸收塔石灰石浆液密度频次分布图

图15 吸收塔壁垢样、健康石膏和循环泵滤网入口垢样的SEM图

参考文献 31

1	孙智滨. 大型燃煤火电机组湿法脱硫系统建模与控制研究[D].上海：上海电力大学, 2019.
	SUN Zhibin. Modeling and control of wet desulfurization system for large coal-fired power plants[D]. Shanghai: Shanghai University of Electric Power, 2019.
2	陈尔鲁. 湿法烟气脱硫过程建模与优化[D]. 杭州：浙江大学, 2016.
	CHEN Erlu. Modeling and optimization of wet flue gas desulfurization process[D]. Hangzhou: Zhejiang University, 2016.
3	张想. 湿法烟气脱硫吸收塔理论建模[D]. 保定：华北电力大学, 2018.
	ZHANG Xiang. Theoretical modeling of wet flue gas desulfurization absorption tower[D]. Baoding: North China Electric Power University, 2018.
4	马双忱, 周权, 曹建宗, 等.湿法脱硫系统动态过程建模与仿真[J].化工学报, 2020, 71(8): 3741-3751.
	MA Shuangchen, ZHOU Quan, CAO Jianzong, et al. Modeling and simulation of wet flue gas desulfurization process[J]. CIESC Journal, 2020, 71(8): 3741-3751.
5	石旻昊. 烟气脱硫过程净烟二氧化硫含量的动态仿真软件的研发[D]. 沈阳：东北大学, 2017.
	SHI Minhao. Development of dynamic simulation software for net sulfur dioxide content in flue gas desulfurization process[D]. Shenyang: Northeastern University, 2017.
6	曹建宗, 林宸雨, 陈文通, 等. 现代预测和优化算法在脱硫系统运行中的应用[J]. 华电技术， 2020， 42(3): 67-71.
	CAO Jianzong, LIN Chenyu, CHEN Wentong, et al. Application of modern prediction and optimization algorithms in FGD systems[J]. Huadian Technology, 2020, 42(3): 67-71.
7	KRZYWANSKI J, CZAKIERT T, BLASZCZUK A, et al. A generalized model of SO₂ emissions from large-and small-scale CFB boilers by artificial neural network approach[J]. Fuel Processing Technology, 2015, 137: 66-74.
8	程换新, 于沙家.面向脱硫系统的BP神经网络预测控制[J]. 石油化工自动化, 2015, 51(5): 39-42.
	CHENG Huanxin, YU Shajia. BP neural network predictive control for desulfurization system[J]. Automation in Petro-Chemical Industry,2015, 51(5): 39-42.
9	孙焰峰. 基于物料平衡的FGD系统性能监测与优化分析[D]. 南京: 东南大学, 2018.
	SUN Yanfeng. Performance monitoring and optimization analysis of FGD system based on material balance[D]. Nanjing: Southeast University, 2018.
10	GUO Y, XU Z, ZHENG C, et al. Modeling and optimization of wet flue gas desulfurization system based on a hybrid modeling method[J]. Journal of the Air & Waste Management Association, 2019, 69(5): 565-575.
11	苏翔鹏, 刁永发, 杨青杰.基于RBF神经网络的双碱法脱硫塔SO₂排放量预测方法[J]. 热力发电, 2017, 46(10): 58-63.
	SU Xiangpeng, DIAO Yongfa, YANG Qingjie. Prediction of SO₂ emission from dual-alkali desulfurization tower based on RBF[J]. Thermal Power Generation, 2017, 46(10): 58-63.
12	苏翔鹏. 基于RBF神经网络的双碱法脱硫塔SO₂排放量预测与优化研究[D]. 上海：东华大学, 2017.
	SU Xiangpeng. Prediction and optimization of SO₂ emissions from double alkali desulfurization tower based on RBF neural network[D]. Shanghai: Donghua University, 2017.
13	李军红, 刘锁清, 董森, 等. 广义回归神经网络对脱硫效率的预测[J]. 自动化技术与应用, 2018, 37(10): 1-3, 37.
	LI Junhong, LIU Suoqing, DONG Sen, et al. Prediction of desulfurization efficiency based on generalized regression neural network[J]. Techniques of Automation and Applications, 2018, 37(10): 1-3, 37.
14	FU J, XIAO H, WANG T, et al. Prediction model of desulfurization efficiency of coal-fired power plants based on LSTM neural network[C]//Proceedings of the 2019 International Conference on Internet of Things (iThings) and IEEE Green Computing and Communications (GreenCom) and IEEE Cyber, Physical and Social Computing (CPSCom) and IEEE Smart Data (SmartData), Atlanta, USA, 2019.
15	张然然. 电厂SO₂排放软测量研究[D]. 济南：齐鲁工业大学, 2019.
	ZHANG Ranran. Research on soft sensing of SO₂ emission in a power plant[D]. Ji’nan: Qilu University of Technology, 2019.
16	PASCANU R, MIKOLOV T, BENGIO Y. On the difficulty of training recurrent neural networks[C]//International Conference on International Conference on Machine Learning. JMLR.org, 2013.
17	HOCHREITER S, SCHMIDHUBER J. Long Short-term memory[J]. Neural Computation, 1997, 9(8): 1735-1780.
18	ALEX Graves. Supervised sequence labelling with recurrent neural networks[M]. Berlin: Springer, 2012.
19	杨国田, 张涛, 王英男, 等. 基于长短期记忆神经网络的火电厂NO_x排放预测模型[J]. 热力发电, 2018, 47(10): 12-17.
	YANG Guotian, ZHANG Tao, WANG Yingnan, et al. Prediction model for NO_x emissions from thermal power plants based on long-short-term memory neural network[J]. Thermal Power Generation, 2018, 47(10): 12-17.
20	张涛. 一种基于长短期记忆神经网络的火电厂NO_x排放量预测方法[D]. 北京：华北电力大学, 2019.
	ZHANG Tao. NO_x emission modeling for power plant boiler based on LSTM[D]. Beijing: North China Electric Power University, 2019.
21	张嘉, 黄景涛, 茅建波. 基于LSTM的电站锅炉NO_x 排放浓度建模研究[J]. 计算机与应用化学, 2018, 35(11): 903-909.
	ZHANG Jia, HUANG Jingtao, MAO Jianbo. NO_x emission modeling for power plant boiler based on LSTM[J]. Computers and Applied Chemistry, 2018, 35(11): 903-909.
22	陈川, 陈冬林, 何李凯. 基于BPNN-EMD-LSTM组合模型的城市短期燃气负荷预测[J]. 安全与环境工程, 2019, 26(1): 153-158, 173.
	CHEN Chuan, CHEN Donglin, HE Likai. Short-term forecast of urban natural gas load base on BPNN-EMF-LSTM combined model[J]. Safety and Environmental Engineering, 2019, 26(1): 153-158, 173.
23	勾玄, 肖先勇. 基于经验模式分解与LSTM神经网络的短期电价预测模型[J]. 西安理工大学学报, 2020, 36(1): 129-134.
	GOU Xuan, XIAO Xianyong. Short-term electricity price forecasting model based on empirical mode decomposition and LSTM neural network[J]. Journal of Xi’an University of Technology, 2020, 36(1): 129-134.
24	陈卓，孙龙祥.基于深度学习LSTM网络的短期电力负荷预测方法[J]. 电子技术, 2018, 47(1): 39-41.
	CHEN Zhuo, SUN Longxiang. Short-term electrical load forecasting based on deep learning LSTM networks[J]. Electronic Technology, 2018, 47(1): 39-41.
25	GLOROT X, BORDES A, BENGIO Y. Deep sparse rectifier neural networks[C]//Proceedings of the 14th International Conference on Artificial Intelligence and Statistics (AISTATS), 2010.
26	DIEDERIK Kingma, JIMMY Ba. Adam: a method for stochastic optimization[C]//ICLR 2015, SanDiego, 2015.
27	SRIVASTAVA N, HINTON G, KRIZHEVSKY A, et al. Dropout: a simple way to prevent neural networks from overfitting[J]. The Journal of Machine Learning Research, 2014, 15(1): 1929-1958.
28	潘海斌, 陈志峰, 王荣. 600MW机组湿法脱硫系统浆液失效的原因分析及解决措施[J]. 电力科技与环保, 2020(1): 34-36.
	PAN Haibin, CHEN Zhifeng, WANG Rong. Cause analysis and solving measures of slurry failure of 600MW unit wet desulfurization system[J].Electric Power Technology and Environmental Protection, 2020(1): 34-36.
29	宫源, 罗安安, 朱家骅, 等. 传质控制下宽分布石膏颗粒群溶解特性[J]. 化工学报, 2018, 69(10): 4177-4183.
	GONG Yuan, LUO An’an, ZHU Jiahua, et al. Dissolution characteristics of widely distributed gypsum particles under mass transfer control[J]. CIESC Journal, 2018,69(10): 4177-4183.
30	潘丹萍, 郭彦鹏, 黄荣廷, 等. 石灰石-石膏法烟气脱硫过程中细颗粒物形成特性[J]. 化工学报, 2015, 66(11): 4618-4625.
	PAN Danping, GUO Yanpeng, HUANG Rongtin, et al. Characteristics of fine particle formation in the process of limestone-gypsum flue gas desulfurization[J]. CIESC Journal, 2015, 66(11): 4618-4625.
31	王子宁, 周加贝, 朱家骅, 等. 二水硫酸钙溶解动力学[J]. 化工学报,2015, 66(3): 1001-1006.
	WUANG Zining, ZHOU Jiabei, ZHU Jiahua, et al. Dissolution kinetics of calcium sulfate dihydrate[J]. CIESC Journal, 2015, 66(3): 1001-1006.

[1]	徐晨阳, 都健, 张磊. 基于图神经网络的化学反应优劣评价[J]. 化工进展, 2023, 42(S1): 205-212.
[2]	吴正浩, 周天航, 蓝兴英, 徐春明. 人工智能驱动化学品创新设计的实践与展望[J]. 化工进展, 2023, 42(8): 3910-3916.
[3]	陈森, 殷鹏远, 杨证禄, 莫一鸣, 崔希利, 锁显, 邢华斌. 功能固体材料智能合成研究进展[J]. 化工进展, 2023, 42(7): 3340-3348.
[4]	李蓝宇, 黄新烨, 王笑楠, 邱彤. 化工科研范式智能化转型的思考与展望[J]. 化工进展, 2023, 42(7): 3325-3330.
[5]	俞俊楠, 俞建峰, 程洋, 齐一搏, 化春键, 蒋毅. 基于深度学习的变宽度浓度梯度芯片性能预测[J]. 化工进展, 2023, 42(7): 3383-3393.
[6]	王俊杰, 潘艳秋, 牛亚宾, 俞路. 分子水平催化重整装置模型构建及应用[J]. 化工进展, 2023, 42(7): 3404-3412.
[7]	王硕, 张亚新, 朱博韬. 基于灰色预测模型的水煤浆输送管道冲蚀磨损寿命预测[J]. 化工进展, 2023, 42(7): 3431-3442.
[8]	赵景斌, 王彦富, 王涛, 马伟恺, 王琛. 基于蒙特卡洛模拟和动态事件树的储罐脆弱性评估[J]. 化工进展, 2023, 42(5): 2751-2759.
[9]	张建伟, 许蕊, 张忠闯, 董鑫, 冯颖. 基于卷积神经网络的撞击流反应器浓度场混合特性[J]. 化工进展, 2023, 42(2): 658-668.
[10]	曾思颖, 杨敏博, 冯霄. 基于机器学习的煤层气组成预测及液化过程的实时优化[J]. 化工进展, 2023, 42(10): 5059-5066.
[11]	王璐, 张磊, 都健. 机器学习高效筛选用于CO₂/N₂选择性吸附分离的沸石材料[J]. 化工进展, 2023, 42(1): 148-158.
[12]	马云飞, 王建兵, 贾超敏, 邢懿心, 柯述, 张先. 臭氧氧化动力学模型及反应器建模研究进展[J]. 化工进展, 2022, 41(S1): 556-570.
[13]	潘艳秋, 李鹏飞, 高石磊, 俞路. 基于MIC筛选规则和BP神经网络的变换装置建模及产品预测[J]. 化工进展, 2022, 41(S1): 36-43.
[14]	李炜, 阮成龙, 王晓明, 李亚洁, 梁成龙. 罐式批次成品汽油调和配方集成建模方法[J]. 化工进展, 2022, 41(9): 4701-4712.
[15]	冯颖, 赵孟杰, 崔倩, 解玉鞠, 张建伟, 董鑫. 分子模拟技术在壳聚糖功能材料开发和应用中的研究进展[J]. 化工进展, 2022, 41(8): 4241-4253.