Search

Machine learning is rapidly becoming a standard technology within the oil and gas industry. This is especially true in petrophysics, where Big Data tend to need more efficient and faster data analysis. The term “machine learning” was coined in 1959 by Arthur Samuel and can be defined as data-driven predictions of behavior rather than rule-based algorithms. Essentially, it is a computer science that uses statistical techniques to give computer systems the ability to learn with data and without being explicitly programmed. A simple example is to record many measurements of the time required for objects of differing attributes to fall various distances and then build a predictive model using linear regression. This predictive model would not be based upon the theory of gravity or the gravitational constant. Instead, through many observations, the model would learn the underlying order in the data. Supplying more data to the model would increase the model’s accuracy. Thus, machine learning models should improve and become better over time as more data become available.

The interpretation of fault surfaces is key to understanding the subsurface geology represented in 3D seismic volumes. The geologic structure represented by seismic reflections can be auto-tracked in the volume. Faults, however, are imaged as discontinuities or changes in curvature in the seismic data. For many years, fault interpretation involved manually picking fault cuts on orthogonal slices through the seismic volume. These fault cuts were grouped into conceptual faults, and 3D fault surfaces were created from the fault cuts.

Many hydrophone-only receiver deghosting approaches assume a stationary free surface profile, or that the data may be represented by linear events within a small spatial aperture. In this paper we propose methods to address these limitations. Firstly, we describe a data-driven methodology to estimate a time-variant free surface profile which may be used in combination with a modified receiver deghosting formulation. Secondly, we propose the use of a tilted hyperboloidal model of the data, which better represents travel-time moveout at short offsets in shallow water regions. Both methods are illustrated on real datasets and demonstrate more effective deghosting than conventional approaches.

The standard processing approach to transform raw shots records into final products suited to geological interpretation involves cascading numerous steps that can be classified in: pre-processing, which aims at correcting the acquisition imperfections or undesired effects (designature, deghosting, geometry), velocity model building which identifies mapping from the data domain to the depth domain and imaging, which applies this mapping. In this paper we show how a Migration Velocity Analysis (MVA) scheme can evolve into a flexible inversion framework that can perform all three steps by inversions. We demonstrate on a 2D real dataset how it handles pre-processing issues and demultiple, as well as providing the velocity model and image domain final products.

A high-resolution regional seismic, gravity and magnetics dataset was acquired offshore southern Gabon, which allowed an in-depth and integrated approach for analyzing the crustal setting in this area. Detailed analysis of the datasets available, and 3D modeling of the crust over this region, suggests that a more complex crustal setting than previously discussed in papers using either long offset 2D seismic, or detailed prospect scale 3D seismic datasets. Placing this information into the framework of the mega-regional publicly available gravity and magnetic datasets provides insights into the possible relationships between the changing upper and lower plate crustal settings experienced at different times in this area.

Most recent depth seismic imaging studies involve both full-wave and ray-based methods as the resolution of complex ill-posed problems often require a wide range of tools. Also ray based methods suffer from well-known drawbacks, they will provide accurate results in most of the cases. Moreover, relying on nonlinear slope tomography, a challenge-driven approach can be designed for each problem by incorporating prior or external information as needed. Here we propose to show such examples of the challenge-driven approach.

In previous publications, a hybrid deblending flow was proposed in which the last step of residual guided noise attenuation using the Noise to Signal Ratio (NSR) map was employed. This flow was proven to be effective and efficient for many OBS surveys. However, due to the more severe semi-coherent cross-talk noise in our data, a modification of the flow is needed. Utilizing the two insights (1) most recorded energy of the seismic data is bound within the direct arrival and shallow reflectors whose two-way-traveltime (TWT) is roughly known and (2) no or minimal remnant noise should be present in the extracted signal; we first focus on retrieving the direct arrival and primaries signals from shallow reflectors followed by retrieving the remaining primaries energy. Each iteration of signal inversion is accompanied by a three-dimensional joint low-rank and sparse inversion (JLSI) noise attenuation to ensure that minimal cross-talk noise would enter into the signal space.

A new non-linear tomographic inversion based method is put forward to resolve large velocity errors associated with complex geology which can cause poor imaging or migration artifacts. We propose to first create a series of trial-velocities from initial velocity by varying the values inside poor imaging zone. Migrations are followed using these trial velocities. The second stage involves CIG picking on these migrated gathers/stacks with tight constraint to ensure reliable picks. These CIG picks are then de-migrated to invariants with their corresponding trial velocities to form a set of invariants as the input to non-linear tomographic inversion.

A joint inversion of P-wave first arrivals, surface wave dispersion curves, and horizon picking of reflectivity image is proposed in order to produce a high resolution Vp (and Vs) model of the near-surface. These three datasets are merged together in a stochastic optimization process through a normalization taking into account of these different domains in the cost function. Resulting model is geologically consistent and reconcile wave velocities and the shallow reflectivity.