The separation of upgoing waves and downgoing waves in vertical seismic profiling is effectively accomplished in the frequency-wavenumber (f-k) domain using a new technique of contour-slice filtering. The response of the filter has no straight edge or boundary and preserves the 'natural state' of the spectra. The cutoff contour level can be reduced as low as desirable to remove background noise and other unwanted signals. In this sense it can also be viewed as a noise-rejection filter. The effectiveness of the process has been clearly demonstrated through synthetic and real examples. Its success with the upgoing waves depends on the degree of coherency of events in the VSP record. To enhance the coherency and to avoid velocity aliasing, we use a novel band-pass frequency filter which is the result of convolving a boxcar spectral window with a frequency-shifted Gaussian function.

Two-dimensional median filters can be designed so that they have properties similar to f-k fan filters. This is done by using the coefficients of a truncated impulse response of an f-k filter as the weight coefficients for the weighted median process. The filter is called a median f-k filter and can be used to discriminate between events on the basis of apparent velocity. The filter appears suitable as a poststack coherency filter because it produces less distortion at wavefield terminations than conventional f-k fan filters. One-dimensional weighted median filters that include negative coefficients are a logical starting point for the analysis of median f-k filters since simple numerical techniques may be used to analyze the behavior of these filters. We show that median filters with negative coefficients do not provide an unbiased estimate of the mean and can misplace the position of steps. Faults on a stacked section may be modeled by steps, and therefore applying a median f-k filter to stacked seismic data could change the position of faults. However, the distortion of steps introduced by median f-k filters is shown to be less than the distortion produced by the corresponding linear f-k filter, and the error in step placement is small. We present simple model examples and a seismic field data example to illustrate differences between linear f-k filters and median f-k filters.

The data obtained by the vertical seismic profiling (VSP) method require unique processing before interpretation. An important step is the separation of upgoing and downgoing signals by application of an f-k velocity filter and muting of the direct wave. Similarly, if the VSP section is transformed to the p-tau plane, the upgoing and downgoing wave fields become separable because they map to different p-tau quadrants according to their dips. This allows either wave field to be reconstructed by application of the inverse Radon transform after windowing in the p-tau plane.Results obtained by the Radon transform wave field separation (RTWS) method are similar to those obtained by f-k velocity filtering. However, by implementing an amplitude-ratio testing filter in the forward slant stack, the quality of the reconstructed wave field is much improved. For a specific range of p-stack values in the forward transform, this filter sorts and removes the unwanted events from the desired ones by testing their consistency in stack amplitudes.Synthetic examples of separating the upgoing wave field from a full VSP wave field by both methods are presented.

Reverse-time migration (RTM) is advantageous for imaging steeply dipping structures, but it may introduce migration smearing artifacts when applied to vertical seismic profiles (VSP). These artifacts mainly result from the limited coverage of VSP acquisition geometry, causing uneven stacking and insufficient noise suppression across multiple shots. Most studies on denoising VSP imaging data focus on limiting the migration aperture and utilizing dip-angle information, while wave equation-based methods often face challenges in accurately determining wavefront propagation directions or rely on prior subsurface structure. Despite recent advances in deep learning, effectively suppressing smearing artifacts remains a significant challenge in VSP RTM. To address this, we propose a novel deep-learning method for suppressing smearing artifacts in VSP RTM using the removal of smearing artifacts residual neural network (RSA-ResNet). Initially, synthetic data are generated based on convolution theory, using data with smearing artifacts as inputs and data without smearing artifacts as labels for the deep-learning network. Subsequently, the RSA-ResNet model is trained to suppress the smearing artifacts associated with VSP RTM. The trained model is then fine-tuned using wavefield decomposition results from target data containing smearing artifacts. Finally, the refined model is applied to VSP RTM data. Experimental results indicate that the waveforms of the reference traces in the two RSA-ResNet-denoised examples closely match the ground-truth solution, effectively demonstrating the superiority of the proposed method over conventional wavefield decomposition methods.

流体因子是储层流体识别中的重要参数,传统的流体因子不能较为准确地识别储层流体,等效流体体积模量对储层流体的变化更加敏感。多波AVO反演是从地震道集中提取等效流体体积模量的重要手段之一。常规的多波AVO反演基于最小二乘或贝叶斯理论,反演精度强烈依赖于初始模型,但大多数实际工区很难建立高精度、高分辨率的初始模型。为了进一步提高储层流体识别的精度,并降低反演对初始模型的依赖程度,在AVO反演理论的指导下,构建基于无监督深度学习的纵波和转换波道集直接反演等效流体体积模量的方法,将该方法应用于X工区实际数据的反演并进行储层流体识别。X工区内井旁道地震道集的试算结果表明,该方法具有较高的精度,反演结果与测井数据的相对误差约为2.949%,绝对误差小于0.1GPa;X工区内反演得到的等效流体体积模量的剖面和时间切片与已知测井解释结果匹配度良好,说明该方法能够较为精确地识别储层流体,具有良好的实用价值。

Fluid factors are key parameters in reservoir fluid identification;however,traditional fluid factors are unable to precisely identify the reservoir fluid.Among these factors,the effective pore fluid modulus is the most sensitive to the fluid.Multi-wave amplitude-versus-offset (MW AVO) inversion is an effective tool for characterizing reservoir fluids.The inverted accuracy of conventional MW AVO inversion methods,which are based on the least squares theory or Bayesian theory,depend heavily on the models that define the initial conditions.However,building such models with high resolution and high precision is difficult in most field areas.To improve the accuracy of reservoir fluid identification and reduce the dependence of the inversion method on the initial models,an unsupervised deep-learning MW AVO inversion method was developed,which incorporates deep learning within a conventional pre-stack AVO.This allows the effective fluid bulk modulus to be inverted directly by using both P-and S-wave gathers.This method was used to perform field data inversion and fluid identification in the X work area and achieved high accuracy.The relative error between the inverted results and the logging data was 2.949%,and the absolute error was less than 0.1 GPa.The inverted profiles and time slices matched well with the log interpretation results,thereby demonstrating that the proposed method can identify reservoir fluid accurately and has a significant practical value.

Accurately upgoing and downgoing wavefield separation is a critical step in vertical seismic profile (VSP) data processing, as its accuracy is directly related to the imaging quality of VSP data. Traditional methods are based mainly on transforms, and their windows are manually set in the transformed domain to obtain the target wavefield. The manual operations often cause errors and affect the accuracy of wavefield separation. In contrast, deep learning algorithms are more automatic, and have achieved a lot in seismic data processing. We propose to employ a conditional generative adversarial network (CGAN) for wavefield separation in VSP data. A CGAN consists of two main components: a generator, which generates new data samples, and a discriminator, which evaluates the generated samples against real data, with both components trained simultaneously in an adversarial manner to improve the quality of generated samples. The full wavefield serves as a constraint to link the generator and discriminator, ensuring that the separated up- and downgoing wavefields align better with the full wavefield. An asymmetric convolution block is introduced to more effectively capture the directional features of the VSP wavefield. To mitigate the influence of amplitude differences between the waves on the network update, the relative downgoing wavefield (obtained by subtracting the predicted upgoing wavefield from the full wavefield) is included in the loss function. Numerical experiments demonstrate that the trained network can effectively learn the characteristics of the up- and downgoing waves, especially their propagation directions, achieving high-precision wavefield separation.