DEEP LEARNING APLICATION FOR GUIDING SEQUENCE BOUNDARY PICKING USING 2-D CONVOLUTIONAL NEURAL NETWORK ARCHITECTURE, CASE STUDY: “BAGALVABAL” FIELD, KUTAI BASIN

• Main Author: Hisyam Tomtowi, Iqbal
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/79184
• Institution: Institut Teknologi Bandung

Deep learning has experienced advancements in recent years for processing large-scale data from various disciplines. It possesses a high level of effectiveness, thereby enhancing quality and reducing the required time compared to previous conventional processes. This phenomenon opens up the potential for utilizing deep learning in the exploration of oil and gas resources in Indonesia using seismic methods. Seismic data typically entails a substantial volume of information, thus deep learning can assist in the interpretation of such data.This research aims to investigate the application of the 2-D Convolutional Neural Network (2-D CNN) architecture in sequence boundary picking to assess its effectiveness in the Kutai Basin context. This study covers four types of architectural models whose number of layers increases with each model. It was found that the model with 5 convolution layers and 4 max pooling layers obtained the smallest RMSE value at 71.7536 ms with confidence level until xline 4872, it was also known that the use of too many layers would reduce the quality of prediction due to overfitting. The prediction results of sequence boundaries using 2-D CNN architecture showed good results without cutting the sequence boundaries marked by discontinuities such as onlap, downlap, and erosional truncation although there were shortcomings such inconsistencies in the prediction results concerning the continuity with well markers