地震研究

针对重大工程结构抗震分析中地震动记录稀缺,以及传统合成方法在物理真实性和多分量适应性上的瓶颈问题,基于日本KiK-net台站近11万条地震动记录,提出了一种物理经验引导型生成对抗网络算子(GM-WGANO)人工地震动合成方法。该方法利用生成对抗网络(GANs)框架,引入傅立叶神经算子(FNO)优化网络结构,结合震级、最小断层距、等效剪切波速、滑动机制和断层构造类别5个物理条件变量,从强震动观测数据中学习地震动的时空特征概率分布,并通过对抗训练生成与真实记录统计特性高度一致的三分量人工时程。结果表明:生成时程在时域上具有与真实记录相近的强震动持时、相位分布及峰值加速度特性; 傅立叶谱与观测数据的误差均小于±1倍标准差; 地震动峰值加速度(PGA)的对数分布均值与观测数据吻合。

To address the scarcity of ground motion records for seismic analysis of major engineering structures and the limitations of traditional synthetic methods in physical consistency and multi-component compatibility,this paper proposes a physically conditioned Generative Adversarial Network(GAN)operator-based method for artificial ground motion synthesis.By integrating Fourier Neural Operators(FNO)into the GAN framework and incorporating five physical conditioning variables—magnitude,fault distance,site shear-wave velocity,slip mechanism,and fault type,the method learns the spatiotemporal characteristics and probabilistic distribution of strong ground motions from recorded data,enabling the generation of high-fidelity three-component accelerograms.Trained on nearly 110 000 ground motion records from Japan's KiK-net stations,the adversely-optimized model produces synthetic time histories statistically indistinguishable from observational data.Results demonstrate that:The synthesized accelerograms exhibit time-domain characteristics closely aligned with real records.Fourier spectra differences remain within one standard deviation of observed data.Logarithmic distributions of peak ground acceleration(PGA)match the observational mean.This approach overcomes the physical representation and multi-component generation constraints inherent in conventional stochastic and hybrid methods,establishing an efficient,scalable,and physically interpretable framework for synthetic ground motion generation in critical engineering applications.

引言
1 理论与方法
2 数据来源及预处理
3 模型生成
4 讨论
5 结论

图1 物理约束型生成对抗网络人工地震动生成模型结构<br/>Fig.1 Artificial earthquake motion generation method using physically constrained generative adversarial network

图1 物理约束型生成对抗网络人工地震动生成模型结构
Fig.1 Artificial earthquake motion generation method using physically constrained generative adversarial network

图2 训练集地震动记录分布<br/>Fig.2 Distribution of seismic motion records in the training set

图2 训练集地震动记录分布
Fig.2 Distribution of seismic motion records in the training set

图3 在不同地震动参数条件下真实与生成三分量地震动时程对比<br/>Fig.3 Comparison of three-component seismic motion time histories under different seismic motion parameter conditions between real seismic motion records and generated seismic motion records

图3 在不同地震动参数条件下真实与生成三分量地震动时程对比
Fig.3 Comparison of three-component seismic motion time histories under different seismic motion parameter conditions between real seismic motion records and generated seismic motion records

图4 训练数据与生成数据的PGA比值的对数值的均值及标准差<br/>Fig.4 Mean value and standard deviation of log(PGAr/PGAf)for training data and generated data

图4 训练数据与生成数据的PGA比值的对数值的均值及标准差
Fig.4 Mean value and standard deviation of log(PGAr/PGAf)for training data and generated data

图5 真实地震动与生成地震动傅立叶均值谱对比<br/>Fig.5 Comparison of the Fourier mean spectra between real seismic ground motions records and generated seismic ground motions

图5 真实地震动与生成地震动傅立叶均值谱对比
Fig.5 Comparison of the Fourier mean spectra between real seismic ground motions records and generated seismic ground motions

Alkhalifah T,Wang H,Ovcharenko O.2022.MLReal:Bridging the gap between training on synthetic data and real data applications in machine learning[J].Artificial Intelligence in Geosciences,3:101-114.
Arjovsky M,Chintala S,Bottou L.2017.Wasserstein generative adversarial networks[C]//Proceedings of the 34th International Conference on Machine Learning-Volume 70.Sydney,Australia:PMLR:214-223.
Bayless J,Abrahamson N A.2019.An empirical model for the interfrequency correlation of epsilon for Fourier amplitude spectra[J].Bulletin of the Seismological Society of America,109(3):1058-1070.
Boore D M.2003.Simulation of ground motion using the stochastic method[J].Pure and Applied Geophysics,160:635-676.
Chen G Y,Li J L,Guo H.2024.Deep generative model conditioned by phase picks for synthesizing labeled seismic waveforms with limited data[J].IEEE Transactions on Geoscience and Remote Sensing,62:1-15.
Esfahani R D D,Cotton F,Ohrnberger M,et al.2023.TFCGAN:Nonstationary ground-motion simulation in the time-frequency domain using conditional generative adversarial network(CGAN)and phase retrieval methods[J].Bulletin of the Seismological Society of America,113(1):453-467.
Florez M A,Caporale M,Buabthong P,et al.2022.Data-driven synthesis of broadband earthquake ground motions using artificial intelligence[J].Bulletin of the Seismological Society of America,112(4):1979-1996.
Frankel A.2009.A constant stress-drop model for producing broadband synthetic seismograms:Comparison with the next generation attenuation relations[J].Bulletin of the Seismological Society of America,99(2A):664-680.
Garcia D,Wald D J,Hearne M G.2012.A global earthquake discrimination scheme to optimize ground-motion prediction equation selection[J].Bulletin of the Seismological Society of America,102(1):185-203.
Gatti F,Clouteau D.2020.Towards blending Physics-based numerical simulations and seismic databases using generative adversarial network[J].Computer Methods in Applied Mechanics and Engineering,372:113421.
Goodfellow I J,Pouget-Abadie J,Mirza M,et al.2014.Generative adversarial nets[C]//Proceedings of the 28th International Conference on Neural Information Processing Systems-Volume.Cambridge,Massachusetts,United States:MIT Press,2672-2680.
Goulet C A,Watson-Lamprey J,Baker J,et al.2008.Assessment of ground motion selection and modification(GMSM)methods for non-linear dynamic analyses of structures[C]//Geotechnical Earthquake Engineering and Soil Dynamics IV.Sacramento,California,United States:American Society of Civil Engineers,1-10.
Graves R W,Pitarka A.2010.Broadband ground-motion simulation using a hybrid approach[J].Bulletin of the Seismological Society of America,100(5A):2095-2123.
Graves R W,Pitarka A.2015.Refinements to the Graves and Pitarka(2010)broadband ground-motion simulation method[J].Seismological Research Letters,86(1):75-80.
Gulrajani I,Ahmed F,Arjovsky M,et al.2017.Improved training of Wasserstein GANs[C]//Proceedings of the 31st International Conference on Neural Information Processing Systems.Red Hook,New York,United States:Curran Associates Inc,5769-5779.
Ho J,Jain A,Abbeel P.2020.Denoising diffusion probabilistic models[C]//Proceedings of the 34th International Conference on Neural Information Processing Systems.Red Hook,New York,United States:Curran Associates Inc,6840-6851.
Huang Y,Yang C,Sun X,et al.2024.Ground-motion simulations using two-dimensional convolution condition adversarial neural network(2D-Cgan)[J].Soil Dynamics and Earthquake Engineering,178:108444.
Kagan Y Y.2002.Seismic moment distribution revisited:I.Statistical results[J].Geophysical Journal International,148(3):520-541.
Katsanos E I,Sextos A G,Manolis G D.2010.Selection of earthquake ground motion records:A state-of-the-art review from a structural engineering perspective[J].Soil Dynamics and Earthquake Engineering,30(4):157-169.
Kingma D P,Welling M.2013.Auto-encoding variational Bayes[J/OL].(2013)[2025-01-12].http://arxiv.org/abs/1312.6114.
Kovachki N,Li Z,Liu B,et al.2023.Neural operator:learning maps between function spaces with applications to PDEs[J].Journal of Machine Learning Research,24(89):1-97.
Li Y M,Ku B H,Zhang S,et al.2020.Seismic data augmentation based on conditional generative adversarial networks[J].Sensors(Basel,Switzerland),20(23):6850.
Luco N,Bazzurro P.2007.Does amplitude scaling of ground motion records result in biased nonlinear structural drift responses?[J].Earthquake Engineering & Structural Dynamics,36(13):1813-1835.
Maechling P J,Silva F,Callaghan S,et al.2015.SCEC broadband platform:system architecture and software implementation[J].Seismological Research Letters,86(1):27-38.
Marano G C,Rosso M M,Aloisio A,et al.2024.Generative adversarial networks review in earthquake-related engineering fields[J].Bulletin of Earthquake Engineering,22(7):3511-3562.
Matinfar M,Khaji N,Ahmadi G.2023.Deep convolutional generative adversarial networks for the generation of numerous artificial spectrum-compatible earthquake accelerograms using a limited number of ground motion records[J].Computer-Aided Civil and Infrastructure Engineering,38(2):225-240.
Rezaeian S,Kiureghian A D.2010.Simulation of synthetic ground motions for specified earthquake and site characteristics[J].Earthquake Engineering & Structural Dynamics,39(10):1155-1180.
Rezaeian S,Kiureghian A D.2012.Simulation of orthogonal horizontal ground motion components for specified earthquake and site characteristics[J].Earthquake Engineering & Structural Dynamics,41(2):335-353.
Rodgers A J,Pitarka A,Pankajakshan R,et al.2020.Regional-scale 3D ground-motion simulations of MW7 Earthquakes on the Hayward Fault,Northern California resolving frequencies 0-10 Hz and including site-response corrections[J].Bulletin of the Seismological Society of America,110(6):2862-2881.
Stafford P J.2017.Interfrequency correlations among Fourier spectral ordinates and implications for stochastic ground-motion simulation[J].Bulletin of the Seismological Society of America,107(6):2774-2791.
Tavakoli B.2005.Empirical-stochastic ground-motion prediction for Eastern North America[J].Bulletin of the Seismological Society of America,95(6):2283-2296.
Withers K B,Olsen K B,Shi Z,et al.2019.Validation of deterministic broadband ground motion and variability from dynamic rupture simulations of buried thrust earthquakes[J].Bulletin of the Seismological Society of America,109(1):212-228.

备注

引言

1 理论与方法

2 数据来源及预处理

3 模型生成

4 讨论

5 结论

期刊信息