Chasing Remaining Potential in Mature East Nile Delta Development Blocks Using Machine Learning

Abstract A growing number of geophysicists and reservoir engineers are being drawn to machine learning due to its many applications in the oil and gas industry, especially in detecting the hydrocarbon accumulations.Direct applications of machine learning currently cover the whole range of tasks related to the interpretation of seismic data, from automatically interpreting faults and horizons to computer-aided seismic well tying, velocity model optimization, geobodies extraction, and seismic facies classification. In this paper, we demonstrate how artificial neural networks can be used to train a computer to identify seismic data and forecast possible gas accumulations as a quick tool for screening remaining potential in development blocks. More than 20 commercial gas accumulations have been discovered in the study area located in Egypt's Nile Deltawith an exploration success rate of more than 60% attributed to most discoveries exhibiting an abnormal AVO seismic behavior. Gas presence in possible prospects was de-risked utilizing traditional pre-stack seismic data analysis and/or seismic facies classification using seismic inversion products. Much time and processing power are required for this work. Machine learning techniques enable a quick and efficient way for screening large volumes of seismic data with astonishing accuracy. Machine learning techniques were employed to enable a quick and efficient way for screening large volumes of seismic data. Training data selection, filtering of the best-discriminating attributes, and monitoring of the learning process were carried out. The neural network was trained on a few examples in the vicinity of wells. The paper shares a successful implementation of a methodology for identifying gas accumulations in mature blocks for the Messinian stratigraphic level, starting from training data selection through filtering of the best-discriminating attributes and monitoring of the learning process until running the predictions. Although the neural network was trained on only a few examples in vicinity of wells, it managed successfully to delineate known accumulations with almost 90% accuracy. The validation of the prediction power increased the belief in the screening outcome of the machine.The newly identified potential gas accumulations were fed into a voxel-based volumetric evaluation workflow to quickly filter out small sub-economic accumulations. The top ranked identified prospects were recommended for further maturation and possible inclusion in future exploration campaigns. The use of artificial neural networks in identifying gas accumulations and screening seismic data is a novel approach. Machine learning techniques offer a quick and efficient solution for analyzing large volumes of seismic data. The successful implementation of this technology in identifying gas accumulations in the mature blocks for the Messinian stratigraphic level is shared in this paper to be used in other areas for identifying potential gas accumulations whenever applicable.