Machine Learning as a Seismic Prior Velocity Model Building Method for Full-Waveform Inversion: A Case Study from Colombia

• DOI: 10.1007/s00024-021-02655-9
• Publication Date: 2021-02-03T12:06:07Z
• Authors: Ursula Iturrarán-...; Andrés M. Muñoz-G...; Octavio Castillo-...; Khemraj Shukla
• Keywords: Petrophysics

Abstract We use machine learning algorithms (artificial neural networks, ANNs) to estimate petrophysical models at seismic scale combining well-log information, data and attributes. The resulting images are the prior inputs in process of full-waveform inversion (FWI). calculate attributes from a stacked reflected 2-D section then train ANNs approximate following parameters: P-wave velocity ( $$V_\mathrm{{p}}$$ <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"><mml:msub><mml:mi>V</mml:mi><mml:mi>p</mml:mi></mml:msub></mml:math> ), density $$\rho $$ xmlns:mml="http://www.w3.org/1998/Math/MathML"><mml:mi>ρ</mml:mi></mml:math> ) volume clay $$V_\mathrm{{clay}}$$ xmlns:mml="http://www.w3.org/1998/Math/MathML"><mml:msub><mml:mi>V</mml:mi><mml:mi>clay</mml:mi></mml:msub></mml:math> ). extend by constraining it with well lithology we establish two classes: sands shales. Consequently, allows us build an initial earth property model which is iteratively refined produce synthetic seismogram that matches observed data. apply 1-D Kennett method as forward modeling tool create , thickness layers (sands or shales) obtained ANNs. A nonlinear least-squares algorithm minimizes residual (or misfit) between data, improves resolution. In order show advantage using ANN for inversion, compare results other models. One these alternative computed via impedance, achieved semblance analysis: root-mean-square (RMS). good agreement when However, poor do not match real acquired log (constant entire section). Nevertheless, improve including layered structure driven (both model. When doing starting estimated there some gain final respect RMS . To assess quality information available wells impedance. This shows benefit employing estimations during obtain geology illustrate computation FWI, provide detailed steps its corresponding GitHub code.