The dynamic and safe operation of underground gas storage (UGS) relies on the accurate prediction of microseismic activity induced by injection and production disturbances.To address the challenges in modeling microseismic time series,including strong nonlinear coupling, lagged effects of external injection/production operations,and multi-scale evolution patterns,this study proposes a microseismic time series prediction model integrating an extended long short-term memory network (xLSTM) with the Informer architecture.In the data processing stage, a multi-source feature set comprising injection/production rates,wellhead pressure,temperature,and other engineering parameters was constructed,and key driving features were selected using Pearson correlation coefficients and mutual information to enhance input representativeness.In the model design, xLSTM was employed to strengthen the modeling of short-term dynamic disturbances and delayed responses,while Informer’s sparse self-attention mechanism and hierarchical distillation encoding structure enabled efficient capture of long-term dependencies and global evolutionary trends.Experimental results demonstrate that the proposed method significantly outperforms benchmark models in predicting microseismic event frequency and energy in terms of mean absolute error (MAE),mean squared error (MSE), and mean absolute percentage error (MAPE).Moreover,when applied to injection/production conditions across different UGS reservoirs within the same block,data validation shows that the model can accurately predict the evolution of microseismic event frequency and energy under complex operational conditions.
BURTONSHAWE J, PALUSZNYA, MOHAMMADPOURA, et al.Effects of mechanical properties of the underburden on induced seismicity along a basement fault during hydrogen storage in a depleted reservoir[J]. iScience, 2024,27(8): 110553 - 110553.
SHAPIROA S, DINSKEC, LANGENBRUCHC, et al.Seismogenic index and magnitude probability of earthquakes induced during reservoir fluid stimulations[J].The Leading Edge,2010,29(3): 304 - 309.
YUP, DEMPSEYD, RINALDIP A, et al.Association between injection and microseismicity in geothermal fields with multiple wells: data‐driven modeling of Rotokawa, New Zealand, and Húsmúli, Iceland[J].Journal of Geophysical Research. Solid Earth,2023,128(4).DOI:10.1029/2022JB025952 .
[12]
QINY, CHENT, MAX, et al. Forecasting induced seismicity in Oklahoma using machine learning methods[J]. Scientific Reports, 2022, 12(1): 9319.
BLOEMHEUVELS, VAN DEN HOOGENJ, JOZINOVIĆD, et al. Graph neural networks for multivariate time series regression with application to seismic data[J]. International Journal of Data Science and Analytics, 2023, 16(3): 317 - 332.
ZHOUY, LIC, ZHANGH, et al. Experimental study on microseismic monitoring of depleted reservoir-type underground gas storage facility in the Jidong Oilfield, North China[J]. Energies, 2025, 18(14): 3762.
[21]
KIRÁLYJ F, OBERHAUSERH.Kernels for Sequentially Ordered Data.[J].Journal of Machine Learning Research, 2019, 20(31): 1-45.
[22]
FATIMAW S S, RAHIMIA.A Review of Time-Series Forecasting Algorithms for Industrial Manufacturing Systems[J].Machines,2024,12(6): 380.
[23]
CHIENH Y S, TUREKJ S, BECKAGEN, et al. Slower is better: revisiting the forgetting mechanism in LSTM for Slower information decay[J]. arXiv preprint arXiv:2021.
[24]
BECKM, PÖPPELK, SPANRINGM, et al. xlstm: Extended long short-term memory[J]. Advances in Neural Information Processing Systems, 2024, 37: 107547 - 107603.
[25]
CHENX, QIUP, ZHUW, et al. Sequence complementor: Complementing transformers for time series forecasting with learnable sequences[C]//Proceedings of the AAAI Conference on Artificial Intelligence. 2025, 39(15): 15913 - 15921.
[26]
ZHOUH, ZHANGS, PENGJ, et al. Informer: Beyond efficient transformer for long sequence time-series forecasting[C] //Proceedings of the AAAI conference on artificial intelligence. 2021, 35(12): 11106 - 11115.
[27]
CHENY, LIC, ZHAOX. TSGformer: A unified temporal-spatial graph transformer with adaptive cross-scale modeling for multivariate time series[J]. Systems, 2025, 13(8): 688.
[28]
LAIG, CHANGW C, YANGY, et al. Modeling long-and short-term temporal patterns with deep neural networks[C]//The 41st international ACM SIGIR conference on research & development in information retrieval. 2018: 95 - 104.