Title : Data analytics for prediction of production for parent- child wells in Montney Formation, Canada
The development of unconventional reservoirs has led to the generation of multitude of data which can be used efficiently to make effective decisions for reservoir management. Data-driven techniques have lower turn-around times and can be used efficiently for production forecasting and developing further reservoir development insights. The current study focuses on using different types of well data such as completion data, production data and, survey data to predict the average gas production per day for shale wells in Montney Formation, Canada using machine learning. Moreover, the study attempts to predict the average gas production of a child well using only the parameters of the parent well. Parent wells are the wells drilled in the reservoir initially and child wells are the infill wells drilled subsequently in further drilling campaigns. The criterion for deciding a child well is based on specific radius threshold, production gap and orientation from the parent well. The most important feature for the predictions is spatial location given by the latitude which has been used as an indicator of the geological and the petrophysical properties. Different models such as Linear Regression, Random Forest, and Xtreme Gradient Boosting were evaluated, and the effort has been to work on interpretable models to gain understanding on the effect of different parameters on average gas production. Moreover, we used dimensionality reduction techniques to reduce training time and reduce multicollinearity in the input parameters. We concluded that using only the parameters of the parent well in the Montney Formation the average gas production of the child well can be predicted with an R-squared value of 0.82. The workflow can be used to guide decisions for drilling child wells and optimize the completion parameters.
Audience Take Away:
- The workflow applied to the Montney Formation shows how data from parent wells can be used to make predictions about future production from child wells. The workflow can be readily transferred to other oil and gas reservoirs.
- The focus has been to make the models interpretable so that we can understand on what factors does the production depend on most. The turnaround time is less than traditional reservoir forecasting techniques so it can be used to get a robust understanding of the parameters affecting production in a specific area quickly.
- The modeling workflow also shows different ways of tuning the hyperparameters of a machine learning model and its implications as well as ways to manipulate features to be used in a model