|Theme||Visible||Selectable||Appearance||Zoom Range (now: 0)|
Abstract Innovations in horizontal drilling and hydraulic fracturing in the last decades have led to exploitation and development of unconventional reservoirs across the globe and particularly in North America. The integration of various disciplines is vital to the success of commercial and economic development of these highly capital intensive projects. This study is intended to evaluate and optimize hydraulic fracture performance of horizontal wells in the Viking Formation, Canada by using machine learning techniques. A Random Forrest Model (RFM) was developed to better understand and quantify the effects of geology, reservoir and completion design parameters on well productivity. For this purpose, first-year oil production (IP365) was used as a well performance indicator. A total of 875 multi-fractured horizontal wells (MFHW) distributed across ten fields were analyzed using descriptive statistics and graphical techniques to establish correlations and patterns, which were used in the development of a random forest model to predict oil production. Finally, model-agnostic methods such as feature importance, partial dependence plots and SHAP values were used to interpret model results. Results indicated that completion length is the most important production driver followed by well geographic location and net pay. On average, wells in the study area with lateral length greater than 750 m tend to have above-average IP365. Optimum proppant intensities and concentrations are seen ranging from 0.37 t/m to 0.49 t/m and 0.5 t/m to 0.58 t/m, respectively. A diminishing effect on production is observed in wells treated with higher values of proppant intensity and concentration. This data-driven methodology is able to identify production drivers at a field and well level by quantifying the effects of each feature on hydrocarbon production. This work will be of importance to exploitation and play development engineers to maximize well productivity and achieve cost efficiencies in the Viking Formation. It will help engineering teams to leverage data science and machine learning techniques in their daily work flow and use it to optimize well completion design.
Abstract Over the past decade, significant supplies of natural gas have been discovered in shale. While the development of new technologies has driven down the cost of gas extraction, pursuing natural gas in shale continues to be risky and capital-intensive. Producers seek the most productive zones in their unconventional basins, as well as continued improvement in hydraulic fracturing processes. Decreasing costs and reducing risk while maximizing gas production necessitates innovative, advanced analytical capabilities that can give you a comprehensive understanding of the reservoir heterogeneity in order to extract hidden predictive information, identify drivers and leading indicators of efficient well production, determine the best intervals for stimulation, and recommend optimum stimulation processes and frequencies. Modeling, simulating and predicting well productivity requires integrated exploratory, predictive and forecasting capabilities underpinned by advanced analytical models to unlock the true potential of each wellbore. Without the critical insight enabled by integrated analysis to pair productivity analysis with economic feasibility, companies face significant risk and uncertainty when developing new wells or optimizing production of extant wellbores. This paper walks through a case study implemented in the Barnett asset in the United States, exemplifying data mining workflows that successfully improved hydrocarbon production. We shall detail analytical methodologies to explicate the optimization of these assets as additional to those workflows expanded upon in SPE paper 149784 presented in Abu Dhabi at the Middle East Unconventional Gas Conference and Exhibition.1