Digital transformation

[DX-02] Neural Network Seismic Inversion Based on Lithofacies Classification and Seismic Attributes to Predict Reservoir Properties Away from Production Wells

*lilik tri hardanto¹, Masako Robb¹ (1. Emerson (Indonesia))

This approach's main objective is to reduce reservoir characterization uncertainties through the optimal incorporation of well log facies and seismic attributes data. The use of neural network seismic inversion algorithms in the inversion workflow estimates reservoir properties far away from production wells. The output volumes of petrophysical properties can be validated using some quality control steps to confirm the prediction accuracy. This research focuses on reservoir sandstone (oil sand) inside a formation. Multiple 3D seismic attributes and data from five wells, including core and petrophysical logs, are used. Lithofacies classes are created utilizing petrophysical logs and available lithologies from the core description. A neural network seismic inversion method is applied for some well locations using petrophysical logs and seismic attributes. Once the predicted petrophysical property volume is obtained from the inversion workflow, volume rendering and geobody detection are applied to detect oil sand distributions within the reservoir zone.

Applying neural network seismic inversion algorithms to predict reservoir distributions away from existing wells can effectively improve reservoir model efficiency compared to conventional modeling. The model showed that reservoir distributions are sensitive to detection intervals. In practical applications, the prediction results are highly consistent with known reservoirs. Obtaining petrophysical properties from seismic data alone suffers from resolution issues. Therefore, the net pay obtained from well data should be used together with the calibration of geobody thickness obtained from neural network machine learning. The neural network seismic inversion algorithm showed the highest precision (93%) validation to actual well data. It can help to develop a field to increase production.