Applying Machine Learning Methods to Search for Missing Net Pay Zones in Mature Oilfields Wells

Abstract

The main aim of this work is to develop new approaches for processes of searching previously missed potential net pay intervals within wells from well log data of brownfields with machine learning algorithms. The practical interest for such a solution is an extraction of additional oil potential recovery from field data in order to prolongate development and production life of mature fields through development of system capable to automatically make recommendations on previously misinterpreted intervals. Besides application of cognitive technologies the developed methodology includes also a workflow of cross-functional interaction of experts from different disciplines (geologist, petrophysicists) between each other and digital system. The system is based on modern deep learning architectures including convolutional and recurrent artificial neural networks. It utilizes well log data and corresponding interpretation of many specialists in order to accumulate or digitize a vast amount of previous experience. It can be used for the following robust re-interpretation of data in those parts of wells which were not previously considered in terms of net pay intervals but have high potential for new oil saturated thicknesses according to geological conditions and previous experience of manual investigation of such intervals. The field test on a basis of described methodology was conducted on assets of Gazpromneft PJSC in Yamanlo-Nenets Autonomous Region, Western Siberia. The comprehensive volume of geological and geophysical information from oilfield was collected including well log data and corresponding results of expert interpretation. This information is used for model training and then predictions about previously uninterpreted intervals are made, providing business user a new interpretation of target geological objects. New interpretation produced by model was compared with current manual interpretation and new net pay intervals were considered as previously missed and potentially oil saturated. At the next step those intervals were examined by petrophysicists, geologists and reservoir engineers in order to estimate probability of oil saturation. Intervals with highest expert marks were proposed for field work and tested by perforation of the target zone. As a result of described process new net pay intervals were found and well, which was suspended, started a new production life. Obtained results confirm high potential of machine learning models application for search of new potential net pay intervals by helping an expert in daily geological and petrophysical tasks.