Completion optimization is performed to determine completion designs that improve well productivity, reduce interference, and enhance return of investment (ROI). The challenges associated with completion optimization include existing depletion of parent wells, well spacing, reservoir geomechanics, and the costs associated with trying multiple solutions in the field. Some companies optimize completions by applying physics-based numerical methods involving fracture and reservoir simulation, while others replicate analogous designs from neighboring wells. In this paper, we present a fast and efficient data-driven solution that determines optimal completion designs and recommends new drilling locations while estimating productivity and returns. The solution works for new wells as well as for DUC wells in an asset.
The machine learning (ML) model was trained based on the random forest algorithm to predict the best 3-month (B3) production. This ML model accounts for the relationships between completion and drilling parameters, existing depletion, and well spacing. The input dataset comprises publicly available data from 40,000 unconventional wells in Texas. After validating the results of the ML model, an optimization algorithm, with an objective function to minimize the difference between target B3 and ML model’s B3, determines the most optimal completion designs. This optimizer, which is a variation of the widely used genetic algorithm, takes target B3, bounds, and places constraints on drilling, completion, and geolocation parameters as inputs and returns the completion design to achieve the target.
The optimizer evaluates thousands of possible solutions in less than 50 s and returns the top five designs. The accuracy of this approach relies heavily on the robustness of the ML model that depends on the availability and quality of the data used to train the model. Recommended completion designs and associated B3 production values were validated against the real production data. It was observed that the ML model captures the local production trends, and the optimizer recommends completion designs aligned with neighboring well data for true vertical depth (TVD), lateral length, azimuth, total proppant, and total fluid pumped. Additionally, the results were validated by predicting the production forecast using the recommended designs and evaluating the economic returns based on the forecast.
We present a novel data-driven ML-based completion optimization approach that overcomes the challenges of traditional costly numerical methods. Through this solution, completion parameters are optimized for drilled but uncompleted (DUC) wells, completion, and drilling parameters are optimized for new locations of interest. New drilling locations are also recommended based on the productivity of the study area. The cloud-native environment of completion optimization provides extensibility through the integration with ML and physics-based solutions to answer the key challenges in the unconventionals. This fully automated approach, which accounts for the underlying trends within major US basins, will reduce screening and decision-making time from months to days and will therefore reduce costs and increase efficiency.
