Role of Statistical Modeling Robust Algorithms for Reservoir Field Development

Abstract Owing to unforeseen drilling problems such as high dragging, tight spots & high pressure kicks, sometimes cancellation of Wireline logging operations becomes essential for securing the drilled wellbores. Such conditions prevent the acquisition of critical wireline data for estimation of reservoir properties for comprehensive reservoir evaluation. This study focuses on ways to improve reservoir quality through applied statistical modeling algorithms using correlated offset acquisition logging data that helps to define the modeling reservoir parameters. Statistical modeling using linear regressions and non-linear regression relationships are the methods for acquiring the best matching outputs from the trained tested offset wells. A non-linear neural network model, as a self-learning system imitating the activity of the human brain, is an extremely powerful method of modeling that can apply extraordinarily difficult relationships and the output showed the results are more robust. In this study, a computationally trained modeled workflow was applied in blind wells that lack sufficient data. The resulting synthetic density derived porosity log was used for effective reservoir characterization and improved interpretation. A modeled density porosity log enabled precise petrophysical reservoir parameters were utilized through an ideal formation evaluation workflow. Lithofacies variations were captured by integrating the radioactive gamma ray, as a shale indicator, for counting the shale volume. Fluid identifying tools like resistivity logs combined with the modeled porosity logs were used for water saturation analysis that helped to define oil-water contact and identify areas of bypassed pay intervals. Reservoir and non-reservoir intervals were separated as sub-zones through integration of synthetic porosity logs. The model delivered petrophysical parameters, such as shale volume, effective porosity, and water saturation, were integrated into the geological model. The results confirmed utilizing the modelled output reservoir parameters, rather than extrapolation, minimizes the uncertainty in geological model, particularly at blind well locations, for oil-in-place calculations. In order to mitigate the challenges of identifying bypassed hydrocarbon intervals, the modeled algorithms present the novel use of forecasting learning methods to evaluate petrophysical parameters for reservoir characterization. The regulated parameters will improve reserve estimates and oil-water contact analysis by reducing the uncertainty for exaggerated or underestimated values. This will result in reducing the uncertainty of deceptive oil-in-place calculation in uncontrolled locations.