From Vision to Validation: Artificial Neural Networks for Novel Compressional and Shear Wave Velocity Models in Iraqi Carbonate Formations

Wave velocity logs are vital in identifying reservoir lithology, physical properties, geomechanical properties, and fluids in geophysical exploration. However, direct measurement of wave velocities through well logs or core labs is often constrained by financial, technical issues, and site conditions. The novelty of this study lies in constructing new models of compressional and shear wave velocities across various scenarios. This paper proposes artificial neural networks and regression analysis techniques to estimate compressional and shear wave velocities. The proposed models were developed based on datasets from 11 wells drilled in the Rumaila oilfield, located in the southern region of Iraq. The raw datasets are composed of compressional wave, shear wave, formation depth, Neutron porosity, and bulk density. The key distinction between this study and previous literature is that it examines all possible instances of data scarcity that may arise in this field of interest, which in turn affects the determination of geomechanical rock properties. The models were developed based on either one parameter, two or three parameters. These models demonstrated a high R2 value of up to 0.974 and a low mean square error as low as 0.0016. A validation has been done using data from other two wells within the same field of interest as well as from one well in Halfaya oil field within Missan governorate in southern Iraq. The results underscored the efficiency and generalizability of the developed ANN models beyond the initial data sets that affirm the reliability in practical fields’ life.