1 College of Artificial Intelligence, China University of Petroleum (Beijing), Beijing 102249, China
2 National Engineering Research Center of Oil and Gas Pipeline Transportation Safety, MOE Key Laboratory of Petroleum Engineering, Beijing Key Laboratory of Urban Oil and Gas Distribution Technology, China University of Petroleum (Beijing), Beijing 102249, China
3 Kunlun Digital Technology Co., Ltd., Beijing 102206, China
4 Oil & Gas Control Center, PipeChina, Beijing 100013, China
Show less
文章历史+
Received
Published
Revised
2025-06-09
2026-06-18
2026-01-07
Online Date
2026-08-04
PDF (13215K)
摘要
天然气需求量预测是天然气生产系统、天然气管道运行管理系统和天然气下游市场管理的一个关键问题。将数据分解算法与时序预测模型结合是提取原始数据特征并提高现有模型预测准确性的常见做法。现有的模型通常直接将分解的序列数据与模型结合,这无法充分发挥分解算法对原始模型的改进作用。提出了一种融合奇异谱分析(singular spectrum analysis,ssa)算法与改进的时间序列混合器(time series mixer,TsMixer)模型的预测模型SSA-TsMixer,该模型创新性地设计了TsMixer模型的特征提取结构,通过对分解后的序列数据在不同特征提取层上的合理分配,实现了序列间的有效交互,确保模型能够最大限度提取分解算法抽象出的统一特征,并抑制噪声数据对预测结果的干扰。构建的模型在12个天然气用户的用气量数据上进行了详尽的实证研究。在所有测试集进行1~14步长的预测时,模型在准确性和稳定性上表现出了更优的性能,平均绝对百分比误差(mean absolute percentage error,MAPE)范围为3.39%~7.57%,原有模型的误差降低了13.65%。此外,在时间步长为1时,通过SSA-TsMixer模型的消融实验验证了所设计的多序列提取特征结构的有效性。SSA-TsMixer模型通过提高预测精度和稳定性,为天然气管网智能调度与控制提供了有效的技术支持,具有显著的工程价值。
Abstract
Natural gas demand forecasting is recognized as a critical issue for natural gas production systems, pipeline management and operational systems, and downstream market management. The combination of data decomposition algorithms with time series forecasting models was commonly adopted to extract features from raw data and improve prediction accuracy. However, existing models usually integrate decomposed sequence data directly, which prevents the full potential of decomposition algorithms from being realized. A novel forecasting model integrating singular spectrum analysis (SSA) with an improved time series mixer (TsMixer), termed SSA-TsMixer, was proposed. The feature extraction structure of the TsMixer model was innovatively redesigned. By allocating decomposed sub-series appropriately across different feature extraction layers, effective interaction among sequences was achieved. This enables the unified features abstracted by the decomposition algorithm to be fully utilized, while the interference of noise data was suppressed. Extensive empirical studies were conducted on gas consumption data from 12 end-users. The model was evaluated on all test sets for forecasting horizons from 1 step to 14 steps. Superior performance in accuracy and stability is demonstrated. The mean absolute percentage error (MAPE) ranges from 3.39% to 7.57%, which corresponds to a 13.65% error reduction compared to the original model. Furthermore, ablation experiments at a 1-step horizon were carried out. The effectiveness of the designed multi-sequence feature extraction structure was verified. The SSA-TsMixer model enhances forecasting accuracy and stability. It provides effective technical support for the intelligent scheduling and control of natural gas pipeline networks, and has significant engineering value.
WU Mian, SONG Shang-fei, LI Meng-yang, YU Wei-chao, WU Ke-ying, SHI Bo-hui, WANG Kai-hong, GONG Jing.
Combining Decomposition Algorithm and Improving TsMixer’s Gas Consumption Prediction Model[J].
Science Technology and Engineering, 2026, 26(17): 7230-7240 DOI:10.12404/j.issn.1671-1815.2504248
CardosoC A V, CruzG L. Forecasting natural gas consumption using ARIMA models and artificial neural networks[J]. IEEE Latin America Transactions, 2016, 14(5): 2233-2238.
ZhangYishuai, LaiHuige, LiYong, et al. Short-term gas load forecasting based on MEA optimized BP neural network[J]. Automation & Instrumentation, 2016, 31(5): 15-19.
[4]
PanekW, WłodekT. Natural gas consumption forecasting based on the variability of external meteorological factors using machine learning algorithms[J]. Energies, 2022, 15(1): 348.
DaiXiaohua, WangYuchun, ZhuYong, et al. Research on load forecasting of natural gas pipeline network based on SVM method[J]. Natural Gas and Oil, 2009, 27(2): 13-15.
[7]
AnagnostisA, PapageorgiouE, DafopoulosV, et al. Applying long short-term memory networks for natural gas demand prediction[C]// 2019 10th International Conference on Information, Intelligence, Systems and Applications (IISA). Patras: IEEE, 2019: 1-7.
[8]
XueG, SongJ, KongX, et al. Prediction of natural gas consumption for city-level DHS based on attention GRU: a case study for a northern Chinese city[J]. IEEE Access, 2019, 7: 130685-130699.
WangKerui, ShaoBilin. Combined prediction of natural gas load based on Time2Vec-LSTM-TCN-Attention[J]. Journal of Nanjing University of Information Science and Technology, 2024, 16(6): 801-809.
[11]
DuJ, ZhengJ, LiangY, et al. A hybrid deep learning framework for predicting daily natural gas consumption[J]. Energy, 2022, 257: 124689.
[12]
XuD, ZhuD.Short-term gas load forecast based on TCN-BiGRU[C]//2021 China Automation Congress (CAC). Beijing: IEEE, 2021: 7978-7983.
[13]
LiuY, GongC, YangL, et al. DSTP-RNN: a dual-stage two-phase attention-based recurrent neural network for long-term and multivariate time series prediction[J]. Expert Systems with Applications, 2020, 143: 113082.
[14]
VaswaniA, ShazeerN, ParmarN, et al. Attention is all you need[C]// Advances in Neural Information Processing Systems. Long Beach: ACM, 2017: 6000-6010.
[15]
ZhouH, ZhangS, PengJ, et al. Informer: beyond efficient transformer for long sequence time-series forecasting[C]// Proceedings of the AAAI Conference on Artificial Intelligence. Online: AAAI, 2021: 11106-11115.
WuH, XuJ, WangJ, et al. Autoformer: decomposition transformers with auto-correlation for long-term series forecasting[J]. 2021. DOI: 10. 48550/arXiv. 2106. 13008.
[18]
WeiN, LiC, DuanJ, et al. Daily natural gas load forecasting based on a hybrid deep learning model[J]. Energies, 2019, 12(2): 218.
[19]
QiaoW, YangZ, KangZ, et al. Short-term natural gas consumption prediction based on Volterra adaptive filter and improved whale optimization algorithm[J]. Engineering Applications of Artificial Intelligence, 2020, 87: 103323.
[20]
LaibO, KhadirM T, MihaylovaL. 21stInternational Conference on Information Fusion ( FUSION).A Gaussian process regression for natural gas consumption prediction based on time series data[C]// Cambridge: IEEE, 2018: 55-61.
XueLiang, GuoRenhui, LiuYang, et al. Vibration-resistant white light interferometry based on non-uniform fast Fourier transform[J]. Acta Optica Sinica, 2024, 44(4): 204-214.
GuanYunan, WangChao, SongBoyang. Acoustic emission signal processing and analysis of wind turbine blades using wavelet analysis[J]. Journal of Xi’an Jiaotong University, 2025, 59(5): 209-216.
YanMengting, TaoXiangming, WangShengjun, et al. Comprehensive evaluation method of hydropower station energy efficiency based on SSA-LSTM model[J]. Water Resources and Power, 2024, 42(2): 177-182.
LiuBin, JiChunlin, CaoLijun, et al. Remaining useful life prediction of lithium-ion batteries based on adaptive noise complete ensemble empirical mode decomposition and BiLSTM-Transformer[J]. Power System Protection and Control, 2024, 52(15): 167-177.
[29]
LiZ, LuoX, LiuM, et al. Wind power prediction based on EEMD-Tent-SSA-LS-SVM[J]. Energy Reports, 2022(8): 3234-3243.
LiuYongfu, ZhangTianying, HuoDianyang, et al. Simultaneous prediction of multiple loads based on singular spectrum analysis and bidirectional LSTM[J]. Science Technology and Engineering, 2025, 25(19): 8099-8107.
[32]
ZengL, LingL, ZhangD, et al. Optimal forecast combination based on PSO-CS approach for daily agricultural future prices forecasting[J]. Applied Soft Computing, 2023, 132: 109833.
[33]
WeiN, LiC, PengX, et al. Daily natural gas consumption forecasting via the application of a novel hybrid model[J]. Applied Energy, 2019, 250: 358-368.
ShaoBilin, RenMeng, TianNing. Hourly natural gas load forecast based on STL-XGBoost-NBEATSx[J]. Journal of China University of Petroleum(Edition of Natural Science), 2024, 48(3): 170-179.
[38]
YangF, LiX, WangM, et al. Waveform: graph enhanced wavelet learning for long sequence forecasting of multivariate time series[C]// Proceedings of the AAAI Conference on Artificial Intelligence. Washington, D. C. : AAAI, 2023: 10754-10761.
[39]
YiK, ZhangQ, FanW, et al. A survey on deep learning based time series analysis with frequency transformation[J]. arXiv prePrint, arXiv: 2302. 02173, 2023.
[40]
CuiZ, QingX, ChaiH, et al. Real-time rainfall-runoff prediction using light gradient boosting machine coupled with singular spectrum analysis[J]. Journal of Hydrology, 2021, 603: 127124.
[41]
WangW, YeF, WangY, et al. A singular spectrum analysis-enhanced BiTCN-selfattention model for runoff prediction[J]. Earth Science Informatics, 2025, 18(1): 31.
[42]
SaneiS, GhodsiM, HassaniH. An adaptive singular spectrum analysis approach to murmur detection from heart sounds[J]. Medical Engineering & Physics, 2011(3): 362-367.
[43]
JaseenaK U, BinsuC K. Decomposition-based hybrid wind speed forecasting model using deep bidi-rectional LSTM networks[J]. Energy Conversion and Management, 2021, 234: 113944.
[44]
HewamalageH, AckermannK, BergmeirC. Forecast evaluation for data scientists: common pitfalls and best practices[J]. Data Mining and Knowledge Discovery, 2023, 37(2): 788-832.
[45]
DuJ, ZhengJ Q, LiangY T, et al. A knowledge-enhanced graph-based temporal-spatial network for natural gas consumption prediction[J]. Energy, 2023. DOI: 10. 1016/J. ENERGY. 2022. 125976.
[46]
WangY, WuH, DongJ, et al. Deep time series models: a comprehensive survey and benchmark[J]. arXiv prePrint, arXiv: 2407. 13278, 2024.
LiuWenlong, TangChaoli, WangZiyu, et al. Improved Informer model based on singular spectrum analysis for regional atmospheric visibility prediction in China[J]. Journal of Chongqing Technology and Business University(Natural Science Edition), 1-8[2026-03-16]. https://kns.cnki.net/kcms/detail/50.1155.N.20240806.1730.002.html.
[49]
WangX, DongS, ZhangR. An integrated time series prediction model based on empirical mode decomposition and two attention mechanisms[J]. Information, 2023, 14(11): 610.