Abstract

Minifrac testing is widely used in the proppant fracturing design workflow. In a minifrac test, different pumping tests yield critical design parameters; a minifrac test requires significant time to perform. Operational efficiency is a key focus in the industry, and recent techniques employing data science can aid in identifying parameter patterns as an alternative to minifrac testing. A novel machine learning algorithm was applied to an existing database to streamline the redesign process from a multistep to a single-step one in conventional reservoirs.

A regression software package was used for this study. Model datasets were constructed based on well locations and reservoirs to reduce the inherent error. The machine learning model used multivariate linear regression. Model inputs included parameters from an injection test with water: closure pressure, transmissibility, reservoir pressure, and fluid efficiency. Outputs were the design parameters typically evaluated from the calibration injection: fluid efficiency with crosslinked gel. The data were preprocessed by removing outliers. Multicollinearity among independent variables was verified to reduce the redundancy of the input variables. The combination of independent variables was then optimized for the highest R-squared value.

The F-test assessment showed that the model was statistically significant. The standard error of the model was insignificant, enabling treatment design with acceptable accuracy. The ratio of training to testing data was 80:20. Five candidate wells were then chosen as a pilot to test and evaluate the model. The wells were geographically located where a high number of offset wells could be used to build the model. An injection test with water was pumped, and the treatment was redesigned based on the prediction of the model using only injection test inputs. The successful placement and the evaluation of the bottomhole pressure behavior during the treatment validated that the design parameters predicted by the statistical model were accurate.

This machine learning technique represents an innovative redesign approach for conventional reservoirs. Implementing the model reduced the complexity of the proppant fracturing treatment redesign process, thus enhancing efficiency, and potentially reduced fracture damage by eliminating minifrac steps with crosslinked gel. Overall operational efficiency was enhanced by 35% to 50%, saving up to 3 days per stage. Stimulation operations in 17 vertical and 21 horizontal multistage wells were designed using this tool, eliminating calibration injection for 140 stages.

You can access this article if you purchase or spend a download.