However, the capabilities of existing fault auto tracking techniques have serious limitations. Some challenges around automatic interpretation are still not resolved: poor data quality, regional and local discontinuity, volume size, reverse or listric faults.
Finally, the algorithm automatically acquires surface groups using the calculated probability field, enabling a user to further customise the fault extraction parameters.
The quality of horizon tracking is one of the key challenges in seismic interpretation.
It is also one of the most time-consuming tasks.
Nowadays there is a large number of auto tracking algorithms available. However, addressing a structural interpretation workflow in a complex geological environment still poses considerable technological limitations. This drive both project timings and costs up.
It provides an effective solution to the problem of auto tracking on different kinds of seismic data, preserving input data specifics (including multiple fault zones and complex horizons).
There are many different filtering techniques available to address various types of noise and to improve the quality of seismic data in general. However, dealing with filter settings is time consuming and often gives insignificant improvements in results. Changing the settings may also distort seismic amplitude and frequency spectrums. Thus, selecting optimal post stack procedures in each specific case is a common challenge that is hard to tackle.
The trained neural network makes it possible to improve the quality of seismic data while maintaining the correct amplitude ratio: amplifying the visual display of fault zones, removing noise, restoring "removed" reflections, smoothing effects, etc.
Accurate salt body interpretation (and other similar stratigraphic objects) usually presents a complex challenge for geoscientists. Extreme dip values, doubtful horizon interpretation (including multi-Z cases), velocity anomalies and poor reflectivity data - all these factors contribute to uncertainties in interpretation.
Geoplat AI machine learning algorithm divides the source data into segments, so the trained convolutional network applies a similar procedure and builds a 3D model.
Traditional multi-attribute analysis techniques combined with RGB blending (or other blending variations) allow to identify geological objects in the seismic volume with a reasonable degree of accuracy. However, this approach has major disadvantages, such as false probabilities and the need to run complex and often
time-consuming attribute calculations.
Based on Geoplat’s technology team experience, this algorithm helps to evaluate areas of interest even in extremely large seismic volumes. It can also be used as a QC tool.