Productivity Index Prediction for Oil Horizontal Wells Using different Artificial Intelligence Techniques

Abstract

Abstract In the production aspect of petroleum engineering, Productivity Index (PI) is considered as a key parameter to develop the inflow performance relationships (IPR). It estimates and forecasts the well productivity and production efficiency. The current practice to obtain PI is to conduct a rate or well test in a producing well after which the PI can be calculated. Many correlations have been developed to predict PI for horizontal oil wells using reservoir and well properties before drilling wells for planning purposes to effectively build and estimate the production system of the well. Artificial intelligence (AI) techniques have been used in the industry to enhance the engineer's ability to forecast and predict the many different and high uncertain outcomes of many of the petroleum industry aspects. AI methods have proven its accuracy for many cases and have been used as successful tools by many oil and gas companies. Prediction/forecasting using AI is a well-known practice and very essential requirement toward a better and more productive industry. This paper discusses and shows the ability of three artificial intelligence methods (Neural Networks, Fuzzy Logic and Functional Networks) to predict productivity index of horizontal oil wells with very good accuracy as compared to several well-known correlations in the industry (Borisov, Giger-Reiss-Jourdan, Renard-Dupuy, Joshi and Economides, Butler and Furui). It discusses, for the first time in the oil industry, the application of the Functional Networks AI techniques in prediction PI of the horizontal oil wells. The models are built using several real field rate tests collected from more than 100 different horizontal oil wells from a field in the Middle East. Also, models showed to overcome the limitations of existing horizontal wells' correlations with high validity of the prediction due the presence of actual field data, not just assumed/simulated data.