AI Assisted Well Performance Prediction Based on Production Profiles Clustering

Abstract

Production forecast is an important part of field development planning using for business planning and economic evaluation of an oil and gas field. The conventional approach to production forecasting includes a bunch of methods from decline curve analysis to hydrodynamic modeling; however, these methods have certain limitations of use. The objective of the study is to investigate the possibility of application AI/ML methods in classification of wells by their historical behavior, and to use this information to predict performance of existing and future wells. The core method of the tool is K-means clustering based on wells production profiles. There are several parameters which values have been taken at the certain time-steps, for example, oil rate, watercut, GOR, flowing bottom-hole pressure, etc. Each parameter's monthly record is combined with reservoir properties to define the input vector for each well. These vectors are assigned to a specific cluster number. The defined clusters can then be assigned to wells with shorter production history in order to predict their future performance. The proposed using pattern recognition of production is useful for the identification of possible neighbor wells influence and tunes production prediction. The method was tested on an oil field in Abu Dhabi with over 50 wells. The results have shown that there is a good predictability for tested dataset. The obtained clusters identified the production performance in existing wells and the production forecast in the nearest future. The method introduces a new approach to wells clustering based of production profiles only. The approach is tested on Abu Dhabi field first time and practical implementation of the method is showcased. The limitation is that it is applicable to wells that have long production history with stable reservoir performance.