作者：武松, 马永光

作者单位：华北电力大学自动化系，河北 保定 071003

关键词：电站锅炉;NO_x排放预测;深度学习;卷积神经网络;双向长短期记忆网络;全局注意力机制;互信息

摘要：

为降低燃煤电站锅炉氮氧化物（NO_x）的排放量，优化燃烧与脱硝控制，提出一种基于数据驱动的融合卷积神经网络（CNN）与双向长短期记忆网络（BiLSTM）的混合深度学习网络模型并引入网络隐层输出全局注意力机制(GAM)来预测锅炉出口NO_x排放量。首先对某2×300 MW电站锅炉运行历史大数据提取并预处理，然后在确定模型输入变量的基础上利用互信息(MI)理论校准各输入变量与NO_x排放量之间的时间延迟并重构样本序列,最后构建CNN-BiLSTM-GAM模型并将该模型与其他几种典型模型的预测效果进行对比。实验结果显示，该模型预测结果的均方根误差（RMSE）、平均绝对百分比误差(MAPE)、决定系数（R-square）分别为：1.866、0.44%、0.983，为各模型中最优值，表明该模型比其他模型有更高的预测精度和更好的泛化能力。

Prediction of NO_x emission from power plant boiler based on mixed deep learning network
WU Song, MA Yongguang
Department of Automation, North China Electric Power University, Baoding 071003, China
Abstract: To reduce nitrogen oxide (NO_x) emissions from coal-fired power plant boiler, optimizing combustion and denitration control, a hybrid deep learning network model based on data-driven fusion of convolutional neural network (CNN) and bidirectional long short memory network (BiLSTM) is proposed, and the network hidden layer output global attention mechanism (GAM) is introduced to predict NO_x emissions from boileroutlet. Firstly, the big data of boiler operation history of a 2×300 MW power station is extracted and preprocessed. Then, on the basis of determining the input variables of the model, the time delay between each input variable and NO_x emission is calibrated by mutual information (MI) theory, and the sample sequence is reconstructed. Finally, the CNN-BiLSTM-GAM model is constructed and the prediction effect of this model is compared with several other typical prediction models. Experimental results show that the root mean square error (RMSE), mean absolute percentage error (MAPE) and determination coefficient (R-Square) of the model are 1.866, 0.44% and 0.983, which are the best among all models, indicating that this model has higher prediction accuracy and better generalization ability than other models.
Keywords: power plant boiler;NO_x emission prediction;deep learning;convolutional neural network;bidirectional long short memory network;global attention mechanism;mutual information
2022, 48(10):166-174  收稿日期: 2022-03-31;收到修改稿日期: 2022-07-27
基金项目: 河北省科技厅重点研发计划（18214523）
作者简介: 武松（1990-），男，河北石家庄市人，硕士研究生，专业方向为发电系统建模、仿真与优化控制
参考文献
[1] 康俊杰. 电站锅炉燃烧和SCR脱硝系统一体化建模与优化控制研究[D]. 北京: 华北电力大学, 2021.
[2] 杨国田, 王英男, 谢锐彪, 等. 基于变量选择的深度置信神经网络锅炉烟气NO _x排放预测[J]. 热力发电, 2020, 49(4): 34-40
[3] 刘菡, 王英男, 李新利, 等. 基于互信息–图卷积神经网络的燃煤电站NO _x排放预测[J]. 中国电机工程学报, 2022, 42(3): 1052-1060
[4] 谢锐彪, 李新利, 王英男, 等. 基于粒子群优化的双向门控循环神经网络燃煤电厂NO _x排放预测[J]. 热力发电, 2021, 50(10): 87-94
[5] LV Y, LV X G, FANG F, et al. Adaptive selective catalytic reduction model development using typical operating data in coal-fired power plants[J]. Energy, 2020, 192: 116589-116601
[6] 朱钰森, 金晓明, 张泉灵. 基于聚类与加权连接的锅炉NO _x 排放量多模型建模[J]. 控制工程, 2019, 26(4): 688-693
[7] TAN P, HE B, ZHANG C, et al. Dynamic modeling of NO _x emission in a 660MW coal-fired boiler with long short-term memory[J]. Energy, 2019, 176: 429-436
[8] 刘岳, 于静, 金秀章. 基于特征优化和改进长短期记忆神经网络的NO _x质量浓度预测[J]. 热力发电, 2021, 50(7): 162-169
[9] LI N, HU Y. The deep convolutional neural network for NO _x emission prediction of a coal-fired boiler[J]. IEEE Access, 2020, 8: 85912-85922
[10] 王英男. 燃煤锅炉的NO_x排放预测与控制研究[D]. 北京: 华北电力大学, 2021.
[11] 李楠. 基于火焰自由基成像的NO_x排放预测算法研究[D]. 北京: 华北电力大学, 2017.
[12] 孙嘉, 张建辉, 卜佑军, 等. 基于CNN-BiLSTM模型的日志异常检测方法[J]. 计算机工程, 2022, 48(7): 151-158,167
[13] 李秉晨, 于惠钧, 丁华轩, 等. 基于CEEMD和 LSTM-ARIMA 的短期风速预测[J]. 中国测试, 2022, 48(2): 163-168
[14] 叶興, 薛家祥. 改进型 LSTM 网络光伏发电功率预测研究[J]. 中国测试, 2019, 45(11): 14-20
[15] 李晓, 卢先领. 基于双重注意力机制和GRU 网络的短期负荷预测模型[J]. 计算机工程, 2022, 48(2): 291-296,305