基于深度学习的多波地震信号智能匹配方法与应用

在油气勘探开发领域,为充分发挥多波多分量地震勘探的技术优势,需进行多波地震信号的高精度匹配处理。异于传统方法通过改变地震信号的传播时间、相位、频率等特征完成匹配,提出了一种基于深度学习的多波地震信号智能匹配方法,它利用了卷积神经网络(CNN)的特征提取能力,直接提取地震信号的波形特征,并辅以重采样抽取转换波(PS)、纵波(PP)和转换波特征损失加权、Adam梯度下降算法更新PS波特征等,使PS波波形在保持整体不变的前提下,在时间域向PP波逼近。通过PP波与PS波的波形匹配,自动完成多波地震信号的传播时间、相位、频率等动力学、运动学和几何学特征匹配。川西坳陷新场3D3C地震资料的应用表明,该方法在多波地震信号的匹配过程中,不需人工干预,具有高精度、高效率、智能化和自动化等优点,匹配后的PS波在保持原始特性的基础上,主频、频宽、波形等更逼近PP波,能有效地描述地层接触关系,更有利于断层识别、地层追踪、岩性边界刻画等地质解释,为多波地质解释、联合反演等奠定坚实基础。

Intelligent matching method based on deep learning for multiwave seismic signals and its application

In the field of oil and gas exploration and development,high-precision matching of multiwave seismic signals needs to be conducted to give full play to the technical advantages of multiwave and multi-component seismic exploration. This research proposes an intelligent matching method based on deep learning for multiwave seismic signals. Different from the traditional method that changes such features of the seismic signal as propagation time,phase,and frequency to complete the matching,this method uses the powerful feature extraction ability of the convolutional neural network (CNN) to directly extract the waveform features of the seismic signal. Moreover,converted wave (PS) extraction by resampling,longitudinal wave (PP) and converted wave feature loss weighting,and the Adam gradient descent algorithm to update PS wave features are also applied so that the waveform of the PS wave approaches that of the PP wave in the time domain with no overall changes. The dynamic,kinematic,and geometric features,such as the propagation time,phase,and frequency,of multiwave seismic signals are matched automatically through the waveform matching between the PP wave and the PS wave. The application of the 3D3C seismic data from Xinchang in the Western Sichuan Depression shows that this method does not require manual intervention in the matching of multiwave seismic signals and that it has the advantages of high precision,high efficiency,intelligence,and automation. In addition to maintaining its original characteristics, the matched PS wave obtains a dominant frequency,bandwidth,and waveform closer to those of the PP wave. Effectively describing the formation contact relationship and being more conducive to geological interpretations,such as fault identification,formation tracking,and lithological boundary chara-〖JP〗cterization,the proposed method lays a solid foundation for subsequent applications such as multiwave contrast geological interpretation and joint inversion.