The Diffusive Time of Flight (DTOF) workflow was used to analyze the completion design, well spacing, and well production for 43 wells completed recently in different landing targets in the Midland Basin. In addition, machine learning was leveraged to identify the key drivers for drainage volume and cumulative oil production after only a short time on production (130 days).
The workflow comprises two steps: (1) The pressures of a group of wells in a similar geological area were measured using downhole gauges. The DTOF workflow was employed to calculate the drainage volume, surface area, and depletion efficiency. (2) A proprietary algorithm was used to calculate the well spacing and track well sequencing. Two machine learning (ML) models were trained, one with DTOF-estimated drainage volume as the target variable, and the other with cumulative production as the target variable. Well spacing, reservoir fluid properties, completion design, and well sequencing information were included as input variables in the ML models. The key drivers for the drainage volumes and cumulative production were identified by the SHapley Additive exPlanations (SHAP) values of the ML models.
The dataset was high-quality because the bottomhole pressure was measured with a downhole gauge. The ML models are shown to have a reasonable accuracy despite the limited number of wells. Such models can identify the nonlinear relationships between variables represented by the corresponding SHAP values. Proppant concentration is shown to be the key driver for drainage volumes and cumulative oil production for this group of wells. A higher initial water saturation is associated with larger drainage volume but lower cumulative oil production. Compared with completion design, initial water saturation, and landing benches, the well spacing and sequencing play a less significant role in completion efficiency and well production.
The diagnostic workflow based on Diffusive Time of Flight was successfully applied to a group of wells in the Midland Basin. It offers quick analytical results even early in a well's life and requires few input parameters. In addition, the analysis makes no assumptions on fracture geometry and flow regimes. Combined with machine learning models and SHAP values, this workflow can assist engineers in conducting look-back analyses and making quick decisions about future well planning.
