Exploring the Potential of Machine Learning Technology to Generate Productivity Maps Using Offset Field Data for a Strategic Coiled Tubing Drilling Program- Onshore Sharjah, UAE

Sharjah National Oil Corporation (SNOC) has operated the Sajaa gas field since the 1980s. From 2003-2006, a highly successful Underbalanced Coiled Tubing Drilling (UBCTD) campaign was executed by drilling 163 multilaterals for production enhancement. SNOC utilized this huge data repository two decades later by performing a data science study. The objective was to improve the understanding of the reservoir and drilling methodologies for an adjacent gas storage field for productivity enhancement. A Machine Learning methodology was used to distribute reservoir model properties contributing to lateral placement planning. Building on the work published in SPE-216458-MS, the team determined critical relationships between various G&G reservoir model and drilling parameters in this second phase of the study. The Sajaa G&G model properties such as porosity, permeability, biostratigraphic zone and saturations were extracted and integrated with the UBCTD legacy dataset, including measured gas rates, productivity index and well trajectories. A machine learning (ML) algorithm was trained on this integrated dataset to predict the measured productivity gained per foot from the G&G model properties at each model cell. The Sajaa-trained ML model was then pointed to the G&G model at the adjacent gas storage field and used to generate a PI/ft map for each biozone of the gas storage field. These maps were used with HCPV maps to define the sweet spot areas to target future drilling in the adjacent gas storage field. The ML model trained on Sajaa showed an 84% accuracy between the measured and predicted PI/ft at Sajaa. This high accuracy increased the confidence in applying the trained model to the gas storage field to generate PI/ft maps at the gas storage field. The outcomes form the basis of the multilateral placement for the UBCTD planned in three development wells to improve the well capacities by over 1.5 times. The data repository and its derived analytics were crucial to visualize all possible relationships between geology, reservoir properties, well productivity and drilling parameters. The study outcomes provided tangible evidence in planning the UBCTD laterals in the gas storage reservoir. The workflow integrating the empirical data and the simulated reservoir properties was vital to utilize the knowledge of the offset field with existing gas-storage model. Phase 1 listed the best-performing biozones which the team integrated with the PI/ft map from Phase 2. SNOC incorporated the outcomes into a customized sweet spot mapping simulation workflow where higher weightage was given to preferable parameters, and multiple realizations of new lateral placement were generated. The paper accentuates the importance of the typically untapped legacy data for well and field development planning. In addition, this paper aims to compare the trained model's properties with the information recorded from the laterals’ drilling campaign. This paper demonstrates the value of data science and analytics on legacy data, where technology reduces drilling and sub-surface uncertainty for a critical development project.