我国以煤炭为主的能源结构，决定了燃煤发电是当前的主要发电方式。煤燃烧产生的氮氧化物（NO_X）是环境污染的主要原因之一，合理优化 NO_X含量是燃煤电厂面临的重要任务。NO_X的传统测量方式存在难以直接测量、维护困难等问题，因此建立灵活高效的 NO_X软测量模型具有重要意义。本研究以陕西某电厂锅炉为背景，以软测量技术理论为主要方法，对燃煤锅炉 NO_X的软测量建模方法进行了深入研究。主要工作如下：
    （1）针对 NO_X软测量模型辅助变量的筛选问题，通过实地考察并分析 NO_X的生成机理，初步筛选机组的锅炉负荷等 22 个变量作为辅助变量，采集运行数据并进行预处理。针对数据的冗余性和相关性复杂的问题，利用主元分析法对辅助变量进行二次分析，精选出主元贡献率高的 17 个变量作为软测量模型的最终辅助变量。
    （2）针对燃煤锅炉数据的时序性问题，提出一种基于长短时记忆网络（LSTM）的NO_X软测量模型。采用平均绝对百分比误差（MAPE）等评价指标分析各模型软测量性能，分别与BP神经网络和极限学习机 ELM建立的NO_X软测量模型分析比较，基于LSTM的 NO_X软测量模型精度较高，MAPE 低至 4.83%。
    （3）针对 LSTM 模型的超参数选取问题，提出采用鲸鱼优化算法（WOA）对学习率、隐含层神经元个数和时间步长进行联合寻优。针对 WOA 易陷入局部最优解的问题，采用一种非线性收敛因子改善其寻优过程，提出一种改进的鲸鱼算法（IWOA）。选取三个测试函数对其进行性能测试，结果表明，IWOA 的平均目标函数值优化了 5.32%。构建基于 IWOA-LSTM 的 NO_X软测量模型。同时以减小 NO_X含量为目标，利用 IWOA在软测量模型的基础上结合变量的边界条件对目标寻优，对可调变量给出优化指导。
       实验结果表明：基于 IWOA-LSTM 的 NO_X 软测量模型测量精度有效提升且泛化性能好，MAPE 小于 3.80%，满足 NO_X的测量需求。同时可调变量的优化指导有助于锅炉燃烧优化控制，对节能减排具有重要意义。
 

       China's coal-based energy structure determines that coal-fired power generation is the main power generation mode. The NO_X produced by coal combustion is one of the main causes of environmental pollution. Rational optimization of NO_X content is an important task facing coal-fired power plants. The traditional measurement method of NO_X has problems such as difficulty in direct measurement and maintenance. So it is of great significance to establish a flexible and efficient NO_X soft measurement model. This research takes the boiler combustion system of a power plant in Shaanxi as the background, and uses soft-sensing technology theory as the main method to conduct an in-depth study on the NO_X soft measurement modeling method of coal-fired boiler. The main work of this research is as follows:

 Aiming at the selection of auxiliary variables of the NO_X soft measurement model, through field investigation and analysis of the NO_X generation mechanism, 22 unit variables such as boiler load are initially selected as auxiliary variables, and operating data is collected and preprocessed. Aiming at the problem of data redundancy and complex relevance, the principal component analysis method is used to perform secondary analysis on auxiliary variables, and 17 variables with high principal component contribution rate are selected as the final auxiliary variables of the soft sensor model.
 Aiming at the sequential problem of coal-fired boiler data, the NO_X soft-sensing model based on long short term memory network (LSTM) is proposed. The average absolute percentage error (MAPE) and other evaluation indicators are used to analyze the soft sensing performance of each model. Compared with the analysis of the NO_X soft sensing model established by the BP neural network and the extreme learning machine ELM, the NO_X soft sensing model based on LSTM has a higher accuracy and the MAPE is as low as 4.83%.
    Aiming at the problem of selecting the hyperparameters of the LSTM model, a whale optimization algorithm (WOA) is proposed to jointly optimize the learning rate, the number of hidden layer neurons and the time step. Aiming at the problem that WOA is easy to fall into the local optimal solution, a nonlinear convergence factor is used to improve its optimization process, the improved whale algorithm (IWOA) is proposed. Three test functions are selected for performance testing. The results show that the average objective function value of IWOA is optimized by 5.32%. Construct a NO_X soft sensor model based on IWOA-LSTM. At the same time, with the goal of reducing the NO_X content, IWOA is used to optimize the target based on the soft-sensing model and the boundary conditions of the variables, and to provide optimization guidance for the adjustable variables.

       The experimental results show that the NO_X soft sensor model based on IWOA-LSTM is easy to optimize and has good generalization performance. The MAPE is less than 3.80%, which meets the measurement requirements of NO_X. At the same time, the optimization guidance of adjustable variables helps to optimize the control of boiler combustion, which is of great significance to energy saving and emission reduction.

[1]Zhao Y, Liu X, Wang S, et al. Energy relations between China and the countries along the Belt and Road: An analysis of the distribution of energy resources and interdependence relationships[J]. Renewable and Sustainable Energy Reviews, 2019, 107: 133-144.

[2]王圣, 孙雪丽, 徐静馨, 等. 我国“十四五”电力发展规划中煤电环境保护要点分析[J]. 电力学报, 2020, 35(02): 143-148+165.

[3]姚明宇, 聂剑平, 张立欣, 等. 燃煤电站锅炉烟气污染物一体化协同治理技术[J]. 热力发电, 2016, 45(03): 8-12.

[4]董博, 刘世宇, 杜忠明, 等. 煤电与雾霾关系研究[J]. 中国电力, 2017, 50(11): 146-151.

[5]He C, Huang G, Liu L, et al. Assessment and offset of the adverse effects induced by PM2.5 from coal-fired power plants in China[J]. Journal of Cleaner Production, 2020, 286: 125397.

[6]李博, 王卫良, 姚宣, 等. 煤电减排对中国大气污染物排放控制的影响研究[J]. 中国电力, 2019, 52(01): 110-117.

[7]Gholami F, MartinTomas, Gholami Z, et al. Technologies for the nitrogen oxides reduction from flue gas: A review[J]. The Science of the Total Environment, 2020, 714(Apr.20): 136712.1-136712.26.

[8]朱晨曦, 郑迎九, 周昊. SNCR+SCR烟气联合脱硝工艺在电站锅炉中的应用[J]. 中国电力, 2016, 49(02): 164-169.

[9]Cheng G, Zhang C. Desulfurization and Denitrification Technologies of Coal-fired Flue Gas[J]. Polish Journal of Environmental Studies, 2018, 27(2): 481-489.

[10]Deng M C, Fujii R. DCS devices based non-linear process control system design for plants with distributed time-delay using particle filter[J]. Journal of Central South University, 2019, 26(12): 3351-3358.

[11]Rossa C, Lehmann T, Sloboda R, et al. A Data-Driven Soft Sensor for Needle Deflection in Heterogeneous Tissue using Just-in-Time Modelling[J]. Medical & Biological Engineering & Computing, 2016, 55(8): 1-14.

[12]林诗洁, 董晨, 陈明志, 等. 新型群智能优化算法综述[J]. 计算机工程与应用, 2018, 54(12): 1-9.

[13]Li X, Niu P, Liu J. Combustion Optimization of a Boiler Based on the Chaos and Lévy Flight Vortex Search Algorithm[J]. Applied Mathematical Modelling, 2018, 58(1): 1-16.

[14]王建明, 刘鑫璐. 基于SAW气敏阵列的火电厂排放物SO2与NO2监测装置设计[J]. 测控技术, 2019, 38(01): 85-88.

[15]张军, 郑成航, 张涌新, 等. 某1000MW燃煤机组超低排放电厂烟气污染物排放测试及其特性分析[J]. 中国电机工程学报, 2016, 36(05): 1310-1314.

[16]Zeng L, Li X, Zhang S, et al. Effects of OFA Ratio on Coal Combustion and NOX Generation of a 600-MW Downfired Boiler after Changing Air Distribution around Fuel-Rich Flow[J]. Journal of Energy Engineering, 2019, 145(1): 04018073. .

[17]Wang Y, Zhou Y. Numerical optimization of the influence of multiple deep air-staged combustion on the NOX emission in an opposed firing utility boiler using lean coal[J]. Fuel, 2020, 269: 116996.

[18]Spijker C, Swaminathan S, Raupenstrauch H. A Numerically Efficient Method for the Prediction of Nitrogen Oxide Emissions in Industrial Furnaces[J]. steel research international, 2020, 91(12): 2000239.

[19]刘长良, 曹威, 王梓齐. 基于MMI-PCA-KLPP二次降维和模糊树模型的NOX浓度软测量方法[J]. 华北电力大学学报(自然科学版), 2020, 47(01): 79-86.

[20]薛美盛, 冀若阳, 王旭. 基于多元线性回归与滚动窗的NOX排放量软测量[J]. 化工自动化及仪表, 2017, 44(08): 721-724.

[21]王春林, 周昊, 李国能, 等. 大型电厂锅炉NOX排放特性的支持向量机模型[J]. 浙江大学学报(工学版), 2006(10): 1787-1791.

[22]印江, 王尚尚, 李丽锋, 等. 基于IPSO-BP算法的CFB锅炉NOX浓度预测[J]. 自动化与仪表, 2021, 36(02): 58-63.

[23]陈琪, 续欣莹, 谢珺, 等. 基于K-ELM和GMM的氮氧化物排放预测[J]. 计算机工程与设计, 2020, 41(01): 231-237.

[24]Li X, Niu P, Li G, et al. An Adaptive Extreme Learning Machine for Modeling NOX Emission of a 300 MW Circulating Fluidized Bed Boiler[J]. Neural Processing Letters, 2017, 46(2): 643-662.

[25]高常乐, 司风琪, 任少君, 等. 基于LSTM的烟气NOX浓度动态软测量模型[J]. 热能动力工程, 2020, 35(03): 98-104.

[26]王芳. 基于智能监测的火电机组节能优化[D]. 太原: 太原理工大学, 2019.

[27]王文广, 赵文杰. 基于GRU神经网络的燃煤电站NOX排放预测模型[J]. 华北电力大学学报(自然科学版), 2020, 47(01): 96-103.

[28]Safdarnejad S M, Tuttle J F, Powell K M. Dynamic modeling and optimization of a coal-fired utility boiler to forecast and minimize NOX and CO emissions simultaneously[J]. Computers & Chemical Engineering, 2019, 124(MAY 8): 62-79.

[29]Surajdeen A. Iliyas, Moustafa. Elshafei, Mohamed A. Habib, et al. RBF neural network inferential sensor for process emission monitoring. 2013, 21(7): 962-970.

[30]Ilamathi P, Selladurai V, Balamurugan K, et al. ANN–GA approach for predictive modeling and optimization of NOX emission in a tangentially fired boiler[J]. Clean Technologies and Environmental Policy, 2013, 15(1): 125-131.

[31]Debiagi P, Nicolai H, Han W, et al. Machine learning for predictive coal combustion CFD simulations-From detailed kinetics to HDMR Reduced-Order models[J]. Fuel, 2020, 274: 117720.

[32]王美琪, 陈恩利, 刘鹏飞, 等. 融合机理与数据的篦冷机温度软测量模型[J]. 仪器仪表学报, 2018, 39(06): 182-188.

[33]李翔宇, 高宪文, 侯延彬. 基于示功图的抽油井动液面软测量机理建模[J]. 控制工程, 2018, 25(03): 464-471.

[34]Hussain F, Wu R S, Wang J X. Comparative study of very short-term flood forecasting using physics-based numerical model and data-driven prediction model[J]. Natural Hazards, 2021: 1-36.

[35]支恩玮, 任密蜂, 程兰, 等. 基于域适应支持向量回归的磨机负荷软测量[J]. 控制工程, 2020, 27(11): 1867-1872.

[36]Maao Ua Ne M, Zouggar S, Krajai G, et al. Modelling industry energy demand using multiple linear regression analysis based on consumed quantity of goods[J]. Energy, 2021, 225(4): 120270.

[37]Guo W, Pan T, Li Z, et al. Model Calibration Method for Soft Sensors using Adaptive Gaussian Process Regression[J]. IEEE Access, 2019, PP(99): 1-1.

[38]Taghavifar H, Taghavifar H, Mardani A , et al. Appraisal of artificial neural networks to the emission analysis and prediction of CO2, soot, and NOx of n-heptane fueled engine[J]. Journal of Cleaner Production, 2016: 1729-1739.

[39]Xu S , R Alturki, Rehman A U , et al. BP Neural Network Combination Prediction for Big Data Enterprise Energy Management System[J]. Mobile Networks and Applications, 2021, 26: 184-190.

[40]毛清华, 赵健博, 李亚周, 等. 基于ELM神经网络的采煤机截割载荷软测量建模方法[J]. 西安科技大学学报, 2020, 40(05): 769-774.

[41]H Zhang, Liu X, Kong X, et al. Stacked Auto-Encoder Modeling of an Ultra-Supercritical Boiler-Turbine System[J]. Energies, 2019, 12(21).

[42]Pan H, Su T, Huang X, et al. LSTM-based soft sensor design for oxygen content of flue gas in coal-fired power plant[J]. Transactions of the Institute of Measurement and Control, 2021, 43(1): 78-87.

[43]Shin H C, Roth H R, Gao M, et al. Deep Convolutional Neural Networks for Computer-Aided Detection: CNN Architectures, Dataset Characteristics and Transfer Learning[J]. IEEE Transactions on Medical Imaging, 2016, 35(5): 1285-1298.

[44]耿志强, 徐猛, 朱群雄, 等. 基于深度学习的复杂化工过程软测量模型研究与应用[J]. 化工学报, 2019, 70(02): 564-571.

[45]Tang Z, Li Y, Kusiak A. A Deep Learning Model for Measuring Oxygen Content of Boiler Flue Gas[J]. IEEE Access, 2020, PP(99): 1-1.

[46]黄新元. 电站锅炉运行与燃烧调整[M]. 中国电力出版社, 2016.

[47]傅永峰. 软测量建模方法研究及其工业应用[D]. 杭州: 浙江大学, 2007.

[48]赵海霞, 周少娜, 肖化. 四种判别粗大误差准则的比较与讨论[J]. 大学物理实验, 2017, 30(05): 105-107+129.

[49]李雅晶, 辛妍丽. 基于SVR的燃煤机组NOX含量的软测量模型[J]. 计算机测量与控制, 2020, 28(08): 62-66.

[50]Wang Y, Zhou Y. Numerical optimization of the influence of multiple deep air-staged combustion on the NOX emission in an opposed firing utility boiler using lean coal[J]. Fuel, 2020, 269: 116996.

[51]丁续达, 金秀章, 张扬. 基于最小二乘支持向量机的改进型在线NOX预测模型[J]. 热力发电, 2019, 48(01): 61-67.

[52]姜子运, 李忠鹏, 李婷婷, 等. 基于改进TSFNN的燃煤电厂SCR入口NOX质量浓度软测量[J]. 热能动力工程, 2020, 35(12): 79-87.

[53]Xie P, Gao M, Zhang H, et al. Dynamic modeling for NOX emission sequence prediction of SCR system outlet based on sequence to sequence long short-term memory network[J]. Energy, 2020, 190: 116482.

[54]张雷. 基于数据驱动的复杂工业过程软测量方法研究与应用[D]. 长沙: 湖南大学, 2019.

[55]Peng T, Cheng Z, Ji X, et al. NOX Emission Model for Coal-Fired Boilers Using Principle Component Analysis and Support Vector Regression[J]. Journal of chemical engineering of Japan, 2016, 49(2): 211-216.

[56]Li J, Yao X, Wang X, et al. Multiscale local features learning based on BP neural network for rolling bearing intelligent fault diagnosis[J]. Measurement, 2019, 153: 107419.

[57]Zhu S, Heddam S, Wu S, et al. Extreme learning machine-based prediction of daily water temperature for rivers[J]. Environmental Earth Sciences, 2019, 78(6): 202.

[58]Greff K, Srivastava R K, J Koutník, et al. LSTM: A Search Space Odyssey[J]. IEEE Transactions on Neural Networks and Learning Systems, 2016, 28(10): 2222-2232.

[59]Zhang D, Lindholm G, Ratnaw Ee Ra H. Use long short-term memory to enhance Internet of Things for combined sewer overflow monitoring[J]. Journal of Hydrology, 2018, 556: 409-418.

[60]Mirjalili, Seyedali, Lewis, et al. The Whale Optimization Algorithm[J]. Advances in engineering software, 2016, 95: 51-67.

[61]Aljarah I, Faris H, Mirjalili S. Optimizing connection weights in neural networks using the whale optimization algorithm[J]. Soft Computing, 2018, 22(1): 1-15.

[62]岳晓宇, 彭显刚, 林俐. 鲸鱼优化支持向量机的短期风电功率预测[J]. 电力系统及其自动化学报, 2020, 32(02): 146-150.

[63]Mrc A, Gqz B, Kdl C. A many-objective population extremal optimization algorithm with an adaptive hybrid mutation operation - ScienceDirect[J]. Information Sciences, 2019, 498: 62-90.

[64]牛培峰, 丁翔, 刘楠, 等. 基于混合鸡群算法和核极端学习机的锅炉NOX排放的预测[J]. 计量学报, 2019, 40(05): 929-936.

[65]Ma Y, Niu P, Yan S, et al. A modified online sequential extreme learning machine for building circulation fluidized bed boiler's NOX emission model[J]. Applied Mathematics & Computation, 2018, 334: 214-226.

[66]徐璁, 王智化, 许岩韦, 等. 优化配风对NOX生成的影响[J]. 燃烧科学与技术, 2016, 22(03): 241-246.