基于融合降噪先验与多尺度特征聚合的微地震震源定位方法OA

With the exponential growth of seismic exploration data,traditional microseismic localization methods struggle to meet the real-time monitoring requirements of hydraulic fracturing operations.Moreover,environmental noise contamination during data acquisition often leads to low signal-to-noise ratios(SNR),which significantly degrade source localization accu-racy.To address these challenges,this study proposes a novel hybrid self-supervised/supervised deep learning framework.First,a convolutional denoising autoencoder(CDAE)is employed for self-supervised pretraining to simultaneously denoise raw seismic data and learn latent waveform features.The encoder component of the CDAE is then repurposed as a feature ex-traction module and cascaded with a fully convolutional localization network to construct an end-to-end joint localization mod-el.Finally,the integrated network is fine-tuned using a limited amount of labeled data to establish a nonlinear mapping from noisy microseismic recordings to spatial source distributions.Validation tests conducted on both linear velocity models and the Marmousi-2 benchmark demonstrate that the proposed framework achieves superior localization accuracy compared with U-Net and other baseline networks,even when trained with minimal labeled data.