基于NBEATS-MARS的飞机燃油流量预测与航空排放计算方法

陈聪; 李豪杰; 师利中; 陈中青

doi:10.13224/j.cnki.jasp.20250218

基于NBEATS-MARS的飞机燃油流量预测与航空排放计算方法

doi: 10.13224/j.cnki.jasp.20250218

陈聪^1,,
李豪杰^1,*, ,,
师利中¹,
陈中青²

1.
中国民航大学航空工程学院，天津 300300
2.
北京飞机维修工程有限公司杭州分公司，杭州 311207

基金项目: 航空科学基金（2024Z061067002）；中央高校基本科研业务费项目（3122020032）

详细信息

作者简介:
陈聪（1982-），女，副教授，硕士，研究方向为航空维修、状态监控与故障诊断。E-mail：cchen@cauc.edu.cn

通讯作者:
李豪杰（2004-），男，研究方向为航空燃效感知与健康管理。E-mail：haojie_li_aero@163.com

中图分类号: V233.7；TP18
计量
- 文章访问数: 303
- HTML浏览量: 181
- PDF量: 33
- 被引次数: 0
出版历程
- 收稿日期: 2025-05-07
- 网络出版日期: 2025-08-06

Aircraft fuel flow prediction based on NBEATS-MARS and calculation methods for aviation emission

1.
College of Aeronautical Engineering，Civil Aviation University of China，Tianjin 300300，China
2.
Hangzhou Branch，Beijing Aircraft Maintenance Engineering Company，Limited，Hangzhou 311207，China

摘要

摘要:
针对传统方法难以准确预测复杂工况下的燃油流量，进而影响排放计算精度的问题，提出一种基于神经网络基函数分解的NBEATS-MARS（neural basis expansion analysis for time series with multi-variable adaptive rapid state-transition）模型。该模型采用多栈分解结构，设计多类型基函数系统，通过基函数分解实现可解释的高精度预测。实验表明：NBEATS-MARS模型方均根误差为59.49，对称平均绝对百分比误差为5.75%，中位数误差仅为0.29%；在爬升巡航下降阶段表现最佳，方均根误差为32.75，对称平均绝对百分比误差为1.87%。基于此构建了综合航空排放计算方法，通过将预测的燃油流量数据作为核心输入，结合发动机排气温度等健康状态参数，实现了二氧化碳、氮氧化物、黑碳和有机碳等多种航空排放物的精确量化。燃油流量预测误差的降低使排放计算不确定性显著减小，巡航阶段排放量计算精度提升至±2%以内。该方法通过提高上游燃油流量预测精度，有效改善了下游航空排放评估的准确性和空间分辨率。
- 燃油流量预测 /
- 航空排放 /
- 多栈基函数分解 /
- 瞬态响应注意力 /
- 发动机排气温度
Abstract:
To address the challenge of accurately predicting fuel flow under complex operating conditions that affects emission calculation precision, a NBEATS-MARS (Neural Basis Expansion Analysis for Time Series with Multi-variable Adaptive Rapid State-transition) model based on neural network basis function decomposition was proposed for aircraft fuel flow prediction. The model employed a multi-stack decomposition structure and designed multiple types of basis function systems to achieve interpretable high-precision prediction through basis function decomposition. Experiments demonstrated that the NBEATS-MARS model achieved a root mean square error of 59.49, a symmetric mean absolute percentage error of 5.75%, and a median error of only 0.29%. The model performed optimally during the climb-cruise-descent phase with a root mean square error of 32.75 and a symmetric mean absolute percentage error of 1.87%. Based on this, a comprehensive aviation emission calculation method was constructed by using predicted fuel flow data as the core input and combining engine exhaust temperature and other health status parameters to achieve precise quantification of various aviation emissions, including carbon dioxide, nitrogen oxides, black carbon, and organic carbon. The reduction in fuel flow prediction error significantly cut down the emission calculation uncertainty, with cruise phase emission calculation accuracy improved to within ±2%. The method effectively enhanced the accuracy and spatial resolution of downstream aviation emission assessment by improving the upstream fuel flow prediction precision.
- fuel flow prediction /
- aviation emissions /
- multi-stack basis function decomposition /
- transient response attention /
- engine exhaust temperature

HTML

图 1 NBEATS-MARS模型整体架构

Figure 1. Overall architecture of NBEATS-MARS model

下载: 全尺寸图片幻灯片

图 2 NBEATS-MARS模型3种类型基函数示意图

Figure 2. Illustration of three types of basis functions in NBEATS-MARS model

下载: 全尺寸图片幻灯片

图 3 航空器飞行剖面中LTO和CCD阶段划分示意图

Figure 3. Schematic diagram of LTO and CCD phase classification in aircraft flight profile

下载: 全尺寸图片幻灯片

图 4 燃油流量预测结果与误差分析

Figure 4. Fuel flow prediction results and error analysis

下载: 全尺寸图片幻灯片

图 5 NBEATS-MARS模型在不同飞行阶段的预测结果及栈分解可视化

Figure 5. Prediction results and stack decomposition visualization of NBEATS-MARS model across different flight phase

下载: 全尺寸图片幻灯片

图 6 不同飞行阶段的特征注意力权重分布

Figure 6. Distribution of characteristic attention weights during different flight stages

下载: 全尺寸图片幻灯片

图 7 某航班全飞行过程中污染物瞬时排放量变化曲线

Figure 7. Variation of instantaneous emission rates of pollutants during a complete flight

下载: 全尺寸图片幻灯片

图 8 不同飞行阶段的主要航空污染物排放量对数分布图

Figure 8. Logarithmic distribution of major aviation pollutants across different flight phases

下载: 全尺寸图片幻灯片

图 9 不同飞行阶段航空排放分布特征

Figure 9. Distribution characteristics of aviation emissions across different flight phases

下载: 全尺寸图片幻灯片

表 1 LTO阶段气态污染物排放指数

Table 1. Gaseous pollutant emission index in the LTO stage

飞行阶段	NOx	CO	HC
起飞	21.57	0.25	0.02
爬升	17.23	0.16	0.02
进近	8.85	3.24	0.05
慢车	4.22	32.07	1.92

下载: 导出CSV

表 2 NBEATS-MARS模型参数配置

Table 2. NBEATS-MARS model parameter configuration

参数名称	数值及说明
基函数维度	[16, 16, 8]
每栈块数量	2
隐藏层大小	128
每块层数	8
激活函数	ReLU
Dropout比率	0.3
输入序列长度	50
批次大小	256
最大训练轮数	200
初始学习率	0.001
优化器	Adam （权重衰减1×10⁻⁵）
梯度裁剪	最大范数为1.0
损失函数	均方误差（MSE）

下载: 导出CSV

表 3 不同时间序列预测模型性能对比

Table 3. Performance comparison of different time series prediction models

模型	RMSE	MAE	sMAPE	中位数误差/%	95%分位误差/%	推理时间/s
LSTM	72.86	58.11	7.71	3.95	18.22	3.74
BiLSTM	65.21	49.97	6.40	3.58	18.51	3.98
TCN	63.20	38.15	5.94	0.36	17.02	3.19
CNN	78.41	54.77	7.04	2.64	19.85	2.61
Transformer	64.23	48.40	5.93	3.46	18.04	3.24
NBEATS	82.62	65.31	8.49	4.62	21.90	2.93
NBEATS-MARS	59.49	42.21	5.75	0.29	17.79	5.36
注：加粗字体表示当前参数下的最优项。

下载: 导出CSV

表 4 NBEATS-MARS模型在不同飞行阶段的性能表现

Table 4. NBEATS-MARS model performance across different flight phases

飞行阶段	RMSE	MAE	sMAPE	中位数误差/%	95%分位误差/%
起飞	66.12	30.18	23.58	0.70	46.08
爬升	95.42	72.87	2.79	1.89	7.28
巡航	32.75	23.33	1.87	0.18	5.87
进近	104.24	89.98	13.54	11.12	26.77
慢车	28.15	16.38	4.23	1.54	14.67

下载: 导出CSV

参考文献(30)

[1]	Eurocontrol. Advanced emission model (AEM) validation report[R]. Brussels, Belgium: Eurocontrol Experimental Centre, 2004.
[2]	JARRY G, MORVAN M, DELAHAYE D. On the generalization properties of deep learning for aircraft fuel flow estimation models [EB/OL]. (2024-01-15) [2024-04-20]. https://arxiv.org/abs/2410.07717.
[3]	陈聪, 候磊, 李乐乐, 等. 基于GRU改进RNN神经网络的飞机燃油流量预测[J]. 科学技术与工程, 2021, 21(27): 11663-11673. CHEN Cong, HOU Lei, LI Lele, et al. Prediction of aircraft fuel flow based on recurrent neural network[J]. Science Technology and Engineering, 2021, 21(27): 11663-11673. (in Chinese CHEN Cong, HOU Lei, LI Lele, et al. Prediction of aircraft fuel flow based on recurrent neural network[J]. Science Technology and Engineering, 2021, 21(27): 11663-11673. (in Chinese)
[4]	KIM Y J, CHOI S, HONG W K. A deep learning approach to flight delay prediction[C]//Proceedings of IEEE International Conference on Big Data. Washington DC: IEEE, 2016: 1880-1889.
[5]	METLEK S. A new proposal for the prediction of an aircraft engine fuel consumption: a novel CNN-BiLSTM deep neural network model[J]. Aircraft Engineering and Aerospace Technology, 2023, 95(5): 838-848. doi: 10.1108/AEAT-05-2022-0132
[6]	段桂英, 姜洪开. 基于数据融合驱动和DLSTM网络的轴承RUL预测[J]. 计算机应用与软件, 2021, 38(12): 22-29. DUAN Guiying, JIANG Hongkai. Bearing RUL prediction based on data fusion drive and DLSTM network[J]. Computer Applications and Software, 2021, 38(12): 22-29. (in Chinese doi: 10.3969/j.issn.1000-386x.2021.12.005 DUAN Guiying, JIANG Hongkai. Bearing RUL prediction based on data fusion drive and DLSTM network[J]. Computer Applications and Software, 2021, 38(12): 22-29. (in Chinese) doi: 10.3969/j.issn.1000-386x.2021.12.005
[7]	ZANG Haipei, ZHU Jinfu, GAO Qiang. Deep learning architecture for flight flow spatiotemporal prediction in airport network[J]. Electronics, 2022, 11(23): 4058. doi: 10.3390/electronics11234058
[8]	FU J, SUN J, LI Y. Fuel consumption prediction of commercial aircraft based on variational mode decomposition[J]. Journal of Aerospace Engineering, 2021, 34(3): 04021014. doi: 10.1061/(ASCE)AS.1943-5525.0001259
[9]	ORESHKIN B N, CARPOV D, CHAPADOS N. N-BEATS: Neural basis expansion analysis for interpretable time series forecasting[C]//Proceedings of International Conference on Learning Representations. Addis Ababa, Ethiopia: ICLR, 2020: 1-22.
[10]	RYERSON M S, HANSEN M, BONN J. Fuel consumption and operational performance[J]. Transportation Research: Part A Policy and Practice, 2015, 74: 249-258.
[11]	HARLASS T, DISCHL R, KAUFMANN S, et al. Measurement report: in-flight and ground-based measurements of nitrogen oxide emissions from latest-generation jet engines and 100% sustainable aviation fuel[J]. Atmospheric Chemistry and Physics, 2024, 24(20): 11807-11822. doi: 10.5194/acp-24-11807-2024
[12]	PAYNTER G C. Overview on fuel flow correlation methods for the calculation of NOx, CO and HC emissions [EB/OL]. (2013-09-15)[2024-04-15]. https://www.researchgate.net/publication/238068899.
[13]	OpenAirlines. The impact of engine wash on exhaust gas temperature (EGT)[EB/OL]. (2024-01-10) [2024-04-18]. https://www.openairlines.com/blog-impact-engine-wash-exhaust-gas-temperature.
[14]	曹惠玲, 徐林, 李玉铭. 性能退化对民机巡航阶段污染物排放的影响研究[J]. 中国民航大学学报, 2022, 42(4): 341-352. CAO Huiling, XU Lin, LI Yuming. Study on the influence of performance degradation on pollutant emission in cruise phase of civil aircraft[J]. Journal of Civil Aviation University of China, 2022, 42(4): 341-352. (in Chinese CAO Huiling, XU Lin, LI Yuming. Study on the influence of performance degradation on pollutant emission in cruise phase of civil aircraft[J]. Journal of Civil Aviation University of China, 2022, 42(4): 341-352. (in Chinese)
[15]	OLIVARES K G, CHALLU C, MARCJASZ G, et al. Neural basis expansion analysis with exogenous variables: Forecasting electricity prices with NBEATSx[J]. International Journal of Forecasting, 2023, 39(2): 884-900. doi: 10.1016/j.ijforecast.2022.03.001
[16]	邵江南, 葛洪伟. 融合残差连接与通道注意力机制的Siamese目标跟踪算法[J]. 计算机辅助设计与图形学学报, 2021, 33(2): 260-269. SHAO Jiangnan, GE Hongwei. Siamese object tracking algorithm combining residual connection and channel attention mechanism[J]. Journal of Computer-Aided Design & Computer Graphics, 2021, 33(2): 260-269. (in Chinese SHAO Jiangnan, GE Hongwei. Siamese object tracking algorithm combining residual connection and channel attention mechanism[J]. Journal of Computer-Aided Design & Computer Graphics, 2021, 33(2): 260-269. (in Chinese)
[17]	BROOMHEAD D S, LOWE D. Radial basis functions, multi-variable functional interpolation and adaptive networks[R]. Malvern, US: Royal Signals and Radar Establishment, 1988.
[18]	LOU Fangyuan, HARRISON H M, KEY N L. Investigation of surge in a transonic centrifugal compressor with vaned diffuser: Part Ⅰ surge signature[J]. Journal of Turbomachinery, 2023, 145(5): 051003. doi: 10.1115/1.4055866
[19]	HSU D, HSU M, GRABENSTATTER H L, et al. Time-frequency analysis using damped-oscillator pseudo-wavelets: application to electrophysiological recordings[J]. Journal of Neuroscience Methods, 2010, 194(1): 179-192. doi: 10.1016/j.jneumeth.2010.09.019
[20]	XU Shenren, YUAN Caijia, HE Chen, et al. Rotating stall inception prediction using an eigenvalue-based global instability analysis method[J]. International Journal of Turbomachinery, Propulsion and Power, 2024, 9(2): 20. doi: 10.3390/ijtpp9020020
[21]	ROY A, PREMCHAND C P, RAGHUNATHAN M, et al. Critical region in the spatiotemporal dynamics of a turbulent thermoacoustic system and smart passive control[J]. Combustion and Flame, 2021, 226: 274-284. doi: 10.1016/j.combustflame.2020.12.018
[22]	CHEN Xiangyi, KOPPE B, LANGE M, et al. Influence of casing groove on rotating instabilities in a low-speed axial compressor[J]. Journal of Turbomachinery, 2023, 145(7): 071015. doi: 10.1115/1.4056863
[23]	GRAPS A. An introduction to wavelets[J]. IEEE Computational Science and Engineering, 1995, 2(2): 50-61. doi: 10.1109/99.388960
[24]	XIN Mai, YE Zhifeng, ZHAO Yu, et al. Comprehensive analysis of aero-engine vibration signals based on wavelet transform method[J]. EURASIP Journal on Advances in Signal Processing, 2023, 2023(1): 117. doi: 10.1186/s13634-023-01079-y
[25]	CIARDIELLO R, et al. The flame expansion process (light-round) during the ignition transient in annular combustors[J]. Combustion and Flame, 2022, 238: 111932. doi: 10.1016/j.combustflame.2021.111932
[26]	Eurocontrol. Eurocontrol method for estimating aviation fuel burnt and emissions[R]. Brussels, Belgium: European Environment Agency, 2016.
[27]	European Union Aviation Safety Agency. ICAO aircraft engine emissions databank [EB/OL]. (2019-06-15) [2024-04-10]. https://www.easa.europa.eu/domains/environment/icao-aircraft-engine-emissions-databank.
[28]	王奕惟, 莫李平, 王奕首, 等. 基于全航段QAR数据和卷积神经网络的航空发动机状态辨识[J]. 航空动力学报, 2021, 36(7): 1556-1563. WANG Yiwei, MO Liping, WANG Yishou, et al. Aero-engine status identification based on full-segment QAR data and convolutional neural network[J]. Journal of Aerospace Power, 2021, 36(7): 1556-1563. (in Chinese WANG Yiwei, MO Liping, WANG Yishou, et al. Aero-engine status identification based on full-segment QAR data and convolutional neural network[J]. Journal of Aerospace Power, 2021, 36(7): 1556-1563. (in Chinese)
[29]	HYNDMAN R J, KOEHLER A B. Another look at measures of forecast accuracy[J]. International Journal of Forecasting, 2006, 22(4): 679-688. doi: 10.1016/j.ijforecast.2006.03.001
[30]	Joint Committee for Guides in Metrology. Evaluation of measurement data-Guide to the expression of uncertainty in measurement: JCGM 100: 2008[R]. Paris: Bureau International des Poids et Mesures, 2008.