Machine Learning Techniques To Assist Natural Fracture Systems Using Coda Wave Interferometry

Abstract:

The quick and correct identification of an efficient fracture network arbitrates the storage in carbonates. Naturally fractured carbonates poise multiple complexities and massive benefits in the cases of water storage or hydrocarbon production. To better understand the fractures, we apply coda wave interferometry (CWI), to record multiple scattered waves on the borehole wall at a limited number of receivers varying time and position. With this technique, we exaggerate the nonlinear dependence of the velocity with fractures, where the temporal change of the velocity in the medium is recorded.

This technique is applied to Mesozoic age rocks in the North of the Chiapas Massive. A dipole sonic tool recorded 662 meters along with a full suite of borehole logs. We calculated the cross-correlation between the unperturbed and perturbed waveforms and the reference waveform. The velocity change from the CWI complements the other datasets to interpret the presence of the fracture network. The increase of (dv/v) corresponds with the presence of natural open fractures. A decrease of (dv/v) ties with cemented fractures, typically cemented with calcite. A stable CWI plot shows small arbitrary differences that shows good correlation with stratigraphy. Despite the noise-like manifestation of the coda portion of the wave, it is highly repeatable yielding a higher precision and level of susceptibility to small structures (approximately similar to the wavelength of the source wavelet ~15kHz). CWI measures the time difference between the waves recorded before travel on a medium with different scattering conditions yielding on a different time lag. Having a different contrast of the CWI behavior might suggest that the open fractures create a reverberation network that are not visible on the direct arrivals but on the coda portion of the waves. On the other hand, sealed fractures tend to have pure mineral crystallization posing a higher velocity path where the coda is transmitted. We used Self-organization Maps to assist fracture classification using multiple datasets trained with the CWI velocity change output.