Abstract:
Traditional geophysical workflows have long been utilized for lithofacies identification and prediction, but the integration of machine learning techniques has become increasingly important to enhance automation and address data quality issues. This study focused on mapping lithofacies in the Lower Goru Formation by establishing relationships between well log petrophysical properties and 3D seismic data using machine learning techniques. It involved building elastic relationships between seismic data, its attributes (sweetness and Impedance) and petrophysical properties at four wells Sawan 1, Sawan 2, Sawan 7, and Sawan 8. The GBR and FNN algorithms were used to predict petrophysical properties and cut-off filters were applied to identify gas sand, wet sand, and shale facies at the C interval horizon. Additionally, Continuous Wavelet Transform (CWT) is applied on seismic data to increase the vertical resolution for the accurate prediction of thin layers of sand and shale. Both models performed well on both datasets, however dataset 2 with the CWT transform on seismic data achieved a coefficient of determination exceeding 97% for all three predicted properties on Sawan 1. GBR was selected as the final model due to its simplicity and slightly better performance than FNN, with R squared values of 97% for VSHL, 98% for PHIE, and 99% for VSHL. The study also developed a web-based application to visualize Segy and GIS data. This research employs the automation of predicting lithofacies and utilization of CWT along with ML algorithms to improve the seismic resolution which could be used in other domains of exploration sector as well.
