In giant reservoirs, production sustainability strongly depends on the identification of opportunities for infill drilling. This paper presents the use of Machine Learning to speed-up and improve the efficiency of the evaluation of future infill wells, in an effort to optimize field development of a Giant Mature reservoir Onshore Abu Dhabi. In the mature giant carbonate reservoir studied, more than 420 wells are already drilled with consistent spacing but with varying orientations. This paper illustrates some examples of settings that are difficult to assess without geometric calculations, leading to time-consuming opportunity identification and classification. The minimum set of input for the program includes existing wells trajectories, faults polygons, contact, and production data. Users can define the minimum drainage area for each well, maturity criteria and drain length. For each subsurface target identified, a polygon and simulation input are generated. The Python program is developed and run on an in-house platform and solve the future wells positioning in three main steps: (1) Geometric screening and identification of locations with required spacing, (2) Analysis of nearby well performance, (3) automatic generation of simulation input for evaluation of the subsurface target.

A multi-billion-barrel oil reservoir at onshore Abu Dhabi has been developed under depletion/aquifer water drive for over 30 years. Recently, significant number of producers have encountered water breakthroughs, some of which are highly suspected to be through the connected fracture network naturally occurred in the reservoir. This premature water breakthrough behavior has hindered the company's mission of increasing and stabilizing the oil production from this reservoir. Predicting future water production, optimizing infill drilling locations and implementing water injection remain as main challenges that the operator has faced. The study described in this paper deployed multiple levels of fracture characterizations in the reservoir simulation model by using unstructured grid to explicitly delineate major/large fractures. The multiple rounds of well-by-well history match has been conducted with focus on water production behavior and pressure. Special attention has been put on the wells showed premature water breakthrough and on the wells identified as "dominated by fractures". The history match process involving many multi-disciplinary work sessions utilized well test data, image logs, seismic data and mud losses to ensure the most meaningful adjustments on fractures descriptions to be implemented. In comparison with the ordinary dual-porosity dual-permeability (DPDP) reservoir simulation model that has been used during the past few years by the operator, this novel approach / new unstructured grid model provides much better fracture descriptions, higher quality history match and more reliable predictions. Its faster run speed may attribute to its adaptive mesh refinement treatment. The history match has shown significant improvements especially at wells that are dominated by flow through fractures. The current forecast exercise focuses on screening of infill locations and ranking of pre-mature water breakthrough risks at locations. Effect of peripheral water injection are also part of the prediction. The ultimate success criterion is reducing/eliminating pre-mature water breakthroughs at infill well locations optimized by this model.