College of Information and Commnication Engineering, Harbin Engineering University, Harbin 150001, China
Show less
文章历史+
Received
Published
Revised
2025-10-27
2026-09-28
2026-01-05
Online Date
2026-10-09
PDF (5194K)
摘要
为解决燃气轮机在恶劣环境下性能参数测量存在传感器部署困难、易损坏的问题,本文采用深度神经网络搭建预测模型,对反映燃气轮机运行效果与运行模式的氮氧化物浓度进行预测,所设计的预测模型结合了两个阶段的处理:第一阶段通过多尺度融合完成整体趋势预测,第二阶段采用小波变换与特征卷积提取完成细节信息预测。实验结果表明:该两阶段模型预测结果的平均绝对误差(mean absolute error, MAE)为0.681,均方根误差(root mean square error, RMSE)为0.995,证明该方法可以有效识别燃气轮机的运行模式以及恶劣环境下的数据波动,能够实现氮氧化物排放浓度的准确预测。
Abstract
In order to address the challenges of sensor deployment and vulnerability in harsh gas turbine environments, a predictive model based on deep neural networks was developed. The model forecasts nitrogen oxide concentrations, which reflect operational performance and modes. A first-stage multi-scale fusion approach was integrated for overall trend prediction. A second-stage design using wavelet transforms and feature convolution was employed for detailed information prediction. The results show that the two-stage model achieves a mean absolute error of 0.681 and a root mean square error of 0.995. This demonstrates the model’s ability to identify operational patterns of gas turbines. It also captures data fluctuations under adverse conditions. Therefore, the model accurately predicts nitrogen oxide emission concentrations.
LI Yun-tai, ZHANG Zhong-min.
Two-stage Prediction Model for Nitrogen Oxides in Gas Turbines Based on Wavelet Transform and Feature Convolution[J].
Science Technology and Engineering, 2026, 26(27): 11822-11829 DOI:10.12404/j.issn.1671-1815.2507459
季节项中较大的周期可以看作是较小周期的聚合。因此,在如图1(a)所示的季节项中,可以使用自下而上的方法,将时间序列的季节项信息向上整合,从而为高尺度的季节项补充信息。与季节项部分相反,在捕捉宏观趋势时,细节变化更加重要。较高尺度的时间序列相比较低尺度,能提供更有用的趋势信息。因此,在如图1(b)所示的趋势项中采用了自上而下的混合方法。上述两种互补方法的结构均为线性层-高斯误差线性单元激活函数(gaussian error linear unit,GeLU)-线性层。
经过小波变换后的特征被输入到简化的Inception中进行特征提取,并结合第一阶段的结果进行线性预测。这一部分由两层线性层组成,并加入了线性整流函数(rectified linear unit, ReLU)和Dropout以提高模型的泛化能力。该阶段模型的主要目的是通过对数据集的二次利用,捕捉并放大数据集中包含的特征信息。通过这种方式,第一阶段的预测结果得到修正,从而获得更准确的预测。融合注意力机制的TimeMixer模型作为第一阶段模型,其与第二阶段模型共享部分参数,包括数据集划分、批量大小和输入序列长度。第二阶段模型独立于第一阶段模型进行训练,两者之间仅保持数据传递关系。在两阶段训练过程中,第二阶段仅使用第一阶段训练中对应当前时间步及历史时间步的输出作为输入,从而严格防止数据泄露。最终模型如图4所示。
MaYuhui, YangYongbo, ZhangNan, et al. Fault diagnosis method for gas turbine fuel system based on measurable data features and regression analysis[J]. Gas Turbine Technology, 2025, 38(3): 43-49.
[3]
SkalskaK, JacekS M, StanislawL. Trends in NOx abatement: a review[J]. Science of the Total Environment, 2010, 408(19): 3976-3989.
[4]
SuraseR S, KonijetiR, ChopadeR P. Thermal performance analysis of gas turbine power plant using soft computing techniques: a review[J]. Engineering Applications of Computational Fluid Mechanics, 2024, 18(1): 1-24.
[5]
RajatS, HanyB, JohnW. A comparison of two trending strategies for gas turbine performance prediction[J]. Journal of Engineering for Gas Turbines and Power, 2008, 130(4): 041601.
[6]
AndreaC, MauroV. Application of forecasting methodologies to predict gas turbine behavior over time[J]. Journal of Engineering for Gas Turbines and Power, 2012, 134(1): 012401.
[7]
VipulG, XuM Y, JayantaK, et al. Prediction enhancement of machine learning using time series modeling in gas turbines[J]. Journal of Engineering for Gas Turbines and Power, 2023, 145(12): 121005.
[8]
ChenZ M, PuY X. Prediction of emissions from gas turbine power generation on GWO-XGBoost-Sobol[J]. Journal of Mechanical Science and Technology, 2024, 38: 1547-1556.
XiangBinbin. GA-SVM based prediction of NOx emissions in gas turbines[J]. Mechanical Management and Development, 2023, 38(2): 47-49, 52.
[11]
RezazadehA. Environmental pollution prediction of NOx by predictive modelling and process analysis in natural gas turbine power plants[J]. Pollution, 2021, 7(2): 481-494.
[12]
KayaH, PinarT, ErdinçU. Predicting CO and NOx emissions from gas turbines: novel data and a benchmark PEMS[J]. Turkish Journal of Electrical Engineering and Computer Sciences, 2019, 27(6): 4783-4796.
LiYaohua, WangJie. Aero-engine life prediction model based on improved temporal convolutional network[J]. Science Technology and Engineering, 2025, 25(31): 13643-13650.
[15]
GuoL J, ZhangS J, HuangQ L. NOx prediction of gas turbine based on dual attention and LSTM[C]// 2022 34th Chinese Control and Decision Conference (CCDC). Hefei: Institute of Electrical and Electronics Engineers Inc, 2022: 4036-4041.
[16]
LiJ J, HanJ, NiuD P, et al. Fast and accurate gas turbine emission prediction based on a light and enhanced Transformer model[J]. Fuel, 2024, 376: 132750.
[17]
SuX, ZhangY M, ZhangY Y, et al. An enhanced multilayer Res-Informer with Savitzky-Golay filter for predicting mixed CO and NOx emissions in gas turbines[J]. Chemometrics and Intelligent Laboratory Systems, 2025, 261: 105379.
[18]
FengN H, LaiS N, YangJ Y, et al. TimeSieve: extracting temporal dynamics through information bottlenecks[J]. arXiv Preprint, arXiv: 2406. 05036, 2024.
ZhongSonghao, LuYi, LiLi, et al. Long term traffic flow forecasting model based on cross period sparse technique[J]. Science Technology and Engineering, 2025, 25(24): 10468-10477.
[21]
WangS Y, WuH X, ShiX M, et al. TimeMixer: decomposable multiscale mixing for time series forecasting[J]. arXiv Preprint, arXiv: 2405. 14616, 2024.
[22]
WuH X, XuJ H, WangJ M, et al. Autoformer: decomposition transformers with auto-correlation for long-term series forecasting[J]. Advances in Neural Information Processing Systems, 2021, 34: 22419-22430.
[23]
SzegedyC, VanhouckeV, IoffeS, et al. Rethinking the inception architecture for computer vision[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Las Vegas: Institute of Electrical and Electronics Engineers Inc, 2016: 2818-2826.
[24]
VaswaniA, ShazeerN, ParmarN, et al. Attention is all you need[C]// Advances in Neural Information Processing Systems. Long Beach: MIT Press, 2017: 1-30.