Neural network seismic inversion based on lithofacies classification and seismic attributes to predict reservoir properties away from production wells

Reservoir characterization uncertainties can be reduced through the optimal incorporation of well log facies and seismic attribute data into seismic inversion workflows. Neural network seismic inversion algorithms in the inversion workflow estimate reservoir properties far away from production wells. Petrophysical properties of the output volumes are validated by several quality control steps to confirm prediction accuracy. This study focuses on reservoir sandstone (oil sand) inside a formation. Multiple 3D seismic attributes and data from five wells, including cores and petrophysical logs, are used as input. Lithofacies classes are created using petrophysical logs and available lithologies from the core description. A neural network seismic inversion method is applied for some well locations. Once the predicted petrophysical property volume is obtained from the inversion workflow, volume rendering and geobody detection are performed to detect oil sand distributions within the reservoir zone. Applying neural network seismic inversion to predict reservoir distributions away from existing wells effectively improves reservoir model efficiency compared to conventional modeling methods. Our model showed that reservoir distributions are sensitive to detection intervals, and the prediction results are highly consistent with existing knowledge of the reservoirs. Because petrophysical properties obtained from seismic data alone can suffer from resolution issues, the net pay from well data should be used together with the calibration of geobody thickness obtained from neural network machine learning. The neural network seismic inversion results showed the highest precision (93%) validation to actual well data. This can help develop a field to increase its production.