This work describes a Shell process for quantifying production scaling factors using best practices in data analytics that has been employed in numerous shale basins including Permian, SCOOP-STACK, Eagle Ford, Haynesville, and the Montney. Shell formed a collaboration with NOVI Labs leveraging their experience with machine learning platforms. This work is a more detailed extension of an earlier work discussing the application for scaling factors to Vaca Muerta wells (McEntyre et al., 2022). We highlight best practices in developing and quality checking data analytics-based models.
Using NOVI Labs’ data analytics platform, Shell developed over 250 machine learning models based on close to 600 wells to scale oil production for a wide range of conditions. Datasets were filtered based on operator, target zone, normalized production, porosity, and completion vintage year. Around 120 distinct input features were examined, such as average porosity, water saturation, total organic content, bulk density, gamma ray, compressional sonic, true vertical depth of lateral, various completion intensity metrics, and various spacing parameter combinations. Results were validated via multiple blind testing in various blocks and landing zones. We document a workflow used to assign log and petrophysical properties when no basin-wide static model is present.
Key insights from our modeling process include:
• A methodology to establish production scaling factors
• Impacts of operating strategy on model results
• Impacts of removing static input, or other key features on prediction capabilities
• How stimulation intensities matter more in lower quality rock
The final model simplifies a non-linear complex solution into a five variable linear equation that allows Shell to confidently build type curves and perform development planning optimization exercises with little to no pilot data. We highlight the applicability of our methodology beyond scaling production to other applications in the oil and gas industry.
The Vaca Muerta play in the Nequeun Basin is undergoing active development, including testing up to five stacked landing zones within a single development unit (Figure 1). Although the well inventory has expanded in recent years, the number of black oil horizontal wells with a year or more of production is still relatively low at fewer than 600 wells. In comparison, a prolific play like the Permian boasts over 32,000 producing horizontal wells. To overcome the limited well data, Shell Argentina has employed software developed by Novi Labs. This software utilizes machine learning (ML) models trained on Vaca Muerta data sets to accurately predict well performance and identify the magnitude of impact of different subsurface, completion, and other well design features. Shell then applies proprietary post-processing on the ML model results to derive production scaling factors. These scaling factors empowered Shell to standardize well performance across various conditions, confidently build type curves, and optimize development planning.
