Research On Artificial Intelligence-based Separation Methodologies And Processing System Development For Vertical Seismic Profiling(VSP) Multi-wave Signals

Seismic exploration technology,centered around seismic signal acquisition,processing,and interpretation,is a pivotal method for oil and gas discovery.One of its efficient and high-precision survey geometry,Vertical Seismic Profiling(VSP),utilizes seismic signal sensors placed in wells to record diverse,multi-path seismic waves rich in subsurface fluid and structural information.This technology is integral for deep subsurface understanding and oil and gas exploration.VSP multi-wave signal separation is categorized into signal-to-noise separation and wave field separation.The latter aims to precisely identify and segregate different types of seismic wave signals in VSP recordings—an essential step in VSP signal processing and a key challenge that has engaged numerous Digital Signal Processing(DSP)experts and exploration geophysicists to address its theoretical and methodological complexities.Recently,the adoption of Distributed Acoustic Sensing(DAS)in VSP has significantly enhanced data acquisition efficiency and scale,albeit introducing challenges due to the increased diversity and complexity of signal and noise types.This dissertation addresses the critical issue of effectively separating complex VSP signals with high fidelity,a pressing need in the domain of high-precision oil and gas exploration and development.It proposes a novel system for intelligent separation and processing of VSP multi-wave signals by integrating artificial intelligence with traditional wave field separation techniques.This approach not only leverages but also supplements existing methodologies where gaps or insufficient research are identified.The key innovations and results of this work include:1.Multi-type Background Noise Suppression: A versatile method combining deep learning and traditional noise suppression techniques optimizes background noise reduction under complex conditions.Techniques like a U-Net with a Global Context and Attention Block(GCB-AB-U-Net)effectively remove strong random noise.Morphological Component Analysis(MCA)addresses single and double frequency interferences,while inversion fitting methods suppress strong coupled noises,significantly enhancing the VSP signal-to-noise ratio.2.Automatic Tracking of Multi-wave VSP Signals: Improved U-NET ++ and Shape Segment Matching Dynamic Time Warping(SSM-DTW)methods facilitate efficient and precise tracking of VSP multi-wave signals along the phase axis.These techniques help overcome challenges related to cross interference among multi-wave signals,providing critical parameters and significantly enhancing the efficiency of signal recognition and wavefield separation.3.High Fidelity Multi-wave Signal Recognition and Wave Field Separation: An automatic recognition method using a Spatial Pyramid Pooling(SPP)U-Net network and an apparent velocity filtering approach based on in-phase axis timelines are developed to address the challenges of multi-wave signal recognition and fidelity in wave field separation.The integration of Singular Value Decomposition(SVD)filtering and automatic gain control ensures amplitude-preserving separation of wave fields.4.Development and Engineering Application of VSP Signal Intelligent Processing Software System: A software system based on C++ and Qt framework,enhanced by deep learning algorithms developed in Python,supports the intelligent separation of VSP multi-wave signals.This system has achieved industrial application standards and has been successfully implemented in CNPC and Shanghai Petroleum and Natural Gas fields,leading to new discoveries in oil and gas.This research aims to enhance the amplitude preservation in signal-to-noise separation,the efficiency of multi-wave phase axis pickup and recognition,and the fidelity of VSP wave field separation,marking a substantial advancement in the engineering application of VSP technology for oil and gas exploration and development.