Structure of an artificial-intelligence-assisted reservoir characterization and field development protocol

Abstract

Characterization of oil reservoirs for effective field development strategies is an arduous task as it requires the integration of the field data collected in various forms (for instance, seismic survey, well log and field historical pressure/production/injection data, etc.) which requires challenging workflows. The implementation of high-fidelity numerical simulators plays a vital role in quantitively identifying the sweet spots for infill drilling and optimizing the field development strategies. However, the structuring of a reliable numerical simulation model requires gridding, property upscaling and history matching, which all could be considered as laborious and time-intensive processes. Moreover, when the availability of the field data is limited, the use of high-fidelity numerical model could be prohibitively difficult, which may necessitate the development of an alternative approach to effectively characterize, integrate and design the field development plans. This paper presents a class of artificial-neural-network based expert systems trained and tested using field data collected from an oil field in North America. The proposed expert system is capable of generating artificial well logs and assessing the hydrocarbon productivity at hypothetical infill drilling locations covered by the seismic survey. Taking advantages of the robust computational efficacies of the expert systems, high-resolution heat maps indicating the productivity sweet spots can be generated to assist the decision makers in selecting the most prolific infill drilling locations when the available field data is insufficient to structure a rigorous numerical simulation model. The expert systems are trained to adapt to the data structure as exhibited by the field exploration and production data and generate realistic predictions.