Colin MacBeth

Effects of learning parameters on learning procedure and performance of a BPNN

We examined the effects of changing learning parameters on the learning procedure and performance of back-propagation neural networks used to pick seismic arrivals. The results show that such change mainly affects the speed of convergence of the learning procedures, and does not affect the BPNN structure and its overall performance. A relationship between the learning parameters and iteration number is obtained. This relationship may be used as a guide to check the convergence of the learning procedure and the BPNN performance. We also use a weight map of BPNN structure to analyze its interior and performance. Two BPNNs used to pick seismic arrivals from three-component and single-component seismograms have similar weight patterns and operate in a similar way, although they have different structures and trained by different training dataset.

The application of back‐propagation neural network to automatic picking seismic arrivals from single‐component recordings

An automatic approach is developed to pick P and S arrivals from single component (1-C) recordings of local earthquake data. In this approach a back propagation neural network (BPNN) accepts a normalized segment (window of 40 samples) of absolute amplitudes from the 1-C recordings as its input pattern, calculating two output values between 0 and 1. The outputs (0,1) or (1,0) correspond to the presence of an arrival or background noise within a moving window. The two outputs form a time series. The P and S arrivals are then retrieved from this series by using a threshold and a local maximum rule. The BPNN is trained by only 10 pairs of P arrivals and background noise segments from the vertical component (V-C) recordings. It can also successfully pick seismic arrivals from the horizontal components (E-W and N-S). Its performance is different for each of the three components due to strong effects of ray path and source position on the seismic waveforms. For the data from two stations of TDP3 seismic network, the success rates are 93%, 89%, and 83% for P arrivals and 75%, 91%, and 87% for S arrivals from the V-C, E-W, and N-S recordings, respectively. The accuracy of the onset times picked from each individual 1-C recording is similar. Adding a constraint on the error to be 10 ms (one sample increment), 66%, 59% and 63% of the P arrivals and 53%, 61%, and 58% of the S arrivals are picked from the V-C, E-W and N-S recordings respectively. Its performance is lower than a similar three-component picking approach but higher than other 1-C picking methods.

Azimuthal variation in P-wave signatures due to fluid flow

The favored dominant mechanism for attenuation in the upper crust at seismic frequencies is intracrack fluid flow. In cracked media, the azimuthal attenuation of the P-wave amplitudes arising from such flow is predicted to be quite substantial. The consequence of this variation in azimuth is a modification in the amplitude behavior of the base event from a cracked reservoir due to transmission through the attenuative layer. Indeed, the effect is of sufficient strength to exacerbate, diminish, or reverse variations that arise solely due to the reflectivity coefficient. Thus, although this attenuation is always greatest perpendicular to the crack strike, the direction of the dimming or brightening of the base reservoir event will depend upon the exact attenuation law and the crack properties. The combination of these factors contributes to a wave behavior that can provide a more adequate discrimination between conditions of brine and oil fill than an interpretation assuming the reflection coefficient alone....

AVD; an emerging new marine technology for reservoir characterization; acquisition and application

Several significant developments in marine technologies in the past few years have resulted in the creation of acquisition techniques suited to azimuthal anisotropy analysis in the offshore environment. The developments have parallelled the evolution in the theory underlying the use of P-P and P-S amplitude versus direction (AVD) for seismic anisotropy estimation. The demands of such AVD methods for a wide azimuthal coverage have only recently been met. To guide future work, the AVD method has been assessed using data from intersecting streamer lines. Application of the method in this example permits an identification of the strike direction of hydrocarbon-filled fractures within a chalk formation in the central North Sea. The results of this study provide confidence that the method is sufficiently sensitive to fractures and can help guide future analyses. The new generation of vertical cables, seabed seismic sensors, and walk-away (and/or 3-D) vertical seismic profiles will eventually lead to high-resolution anisotropy estimation in the offshore environment using this approach.

Application of back-propagation neural networks to identification of seismic arrival types

A back-propagation neural network (BPNN) approach is developed to identify P- and S-arrivals from three-component recordings of local earthquake data. The BPNN is trained by selecting trace segments of P- and S-waves and noise bursts converted into an attribute space based on the degree of polarization (DOP). After training, the network can automatically identify the type of arrival on earthquake recordings. Compared with manual analysis, a BPNN trained with nine groups of DOP segments can correctly identify 82.3% of the P-arrivals and 62.6% of the S-arrivals from one seismic station, and when trained with five groups from a training dataset selected from another seismic station, it can correctly identify 76.6% of the P-arrivals and 60.5% of S-arrivals. This approach is adaptive and needs only the onset time of arrivals as input, although its performance cannot be improved by simply adding more training datasets due to the complexity of DOP patterns. Our experience suggests that other information or another network may be necessary to improve its performance.

Interpreting data matrix asymmetry in near-offset, shear-wave VSP data

Poor experimental control in shear‐wave VSPs may contribute to unreliable estimates of shear‐wave splitting and possible misinterpretation of the medium anisotropy. To avoid this, the acquisition and processing of multicomponent shear‐wave data needs special care and attention. Measurement of asymmetry in the recorded data matrix using singular‐value decomposition (SVD) provides a useful way of examining possible acquisition inaccuracies and may help guide data conditioning and interpretation to ensure more reliable estimates of shear‐wave polarization azimuth. Three examples demonstrate how variations in shear‐wave polarization and acquisition inaccuracies affect the SVD results in different ways. In the first example, analysis of synthetic seismograms with known depth changes in the polarization azimuth show how these may be detected. In the second example, a known source re‐orientation and polarity reversal is detected by applying SVD to near‐offset, shear‐wave VSP data, recorded in the Romashkino fiel...

Data-matrix asymmetry and polarization changes from multicomponent surface seismics

We present methods for interpreting data‐matrix asymmetry and polarization changes with depth from multicomponent surface seismics. There are two main sources of data matrix asymmetry in four component shear‐wave seismics: that arising from the acquisition geometry caused by source and receiver misorientation, misalignment, imbalance, and cross‐coupling, and that arising from the medium caused by variations in the geological structure, lithology, or stress. The asymmetry caused by acquisition geometry is more significant than that from the medium. Two asymmetry indices are used to quantify these medium and acquisition asymmetries separately. Their behavior may be used to identify the origin of the asymmetry. The asymmetry caused by the medium is studied by deriving approximate normal‐incidence, plane‐wave reflection coefficients for an interface separating two anisotropic media with differenfly oriented symmetry axes. The degree of asymmetry in the reflectivity is proportional to the product of the degree...

A comparison of global optimisation methods for near-offset VSP inversion

Abstract Three global optimisation algorithms are applied to the problem of geophysical inversion. We describe and test the methods of Tabu Search, Genetic Algorithms and Simulated Annealing. These techniques are used to invert observations of shear-wave splitting from near-offset Vertical Seismic Profiles. Each search shows distinct advantages and disadvantages so that no particular algorithm can be clearly recommended. Nonetheless, we can recommend that a global optimisation be followed by a local search.

Processing of a nine‐component near‐offset VSP for seismic anisotropy

A convolutional sequence of matrix operators is offered as a convenient deterministic scheme for processing a multicomponent vertical seismic profile (VSP). This sequence is applied to a nine-component near-offset VSP recorded at the Conoco borehole test facility, Kay County, Oklahoma. These data are corrected for tool spin and near-surface anisotropy together with source coupling or imbalance. After wave-field separation using a standard f-k filter, each source and receiver pair for the upgoing waves is adjusted to a common reference depth using a matrix operator based on the downgoing wavefield. The up- and downgoing waves are then processed for anisotropy by a similarity transformation, to separate the qS1 and qS2 waves, from which the anisotropic properties are estimated. These estimates reveal a strong (apparent) vertical birefringence in the near-surface, but weak or moderate values for the majority of the subsurface. The target zone consists of a thin sandstone and deeper shale layer, both of which possess a strong vertical birefringence. The sandstone corresponds to a zone of known fluid flow. An observed qS2 attenuation and polarization change in the shale suggest it contains large fractures.

Separation of Up and Down-Going Wave Fields in Vertical Cable Seismic

Vertical cable seismic (VCS) represents a new method to acquire 3-D marine data by using vertical arrays of hydrophones, in which both up and down-going wave fields are recorded. Despite analogies that may be drawn with VSP, VCS data processing does in fact require a different set of techniques for the separation of these wave fields due to its unique acquisition characteristics. In this study, a filtering method based on solving differential equations in the time and space domain, is used in a procedure to separate up and down-going wave fields, directly after the initial step of alignment. Synthetic tests show that compared to median filtering, or the f k filtering approach, this method obtains the best result. Furthermore, application to real VCS data indicates an encouraging performance.