Application of Machine Learning to Predict Estimated Ultimate Recovery for Multistage Hydraulically Fractured Wells in Niobrara Shale Formation

The completion design of multistage hydraulic fractured wells including the cluster spacing injected proppant and slurry volumes has shown a great influence on the well production rates and estimated ultimate recovery (EUR). EUR estimation is a critical process to evaluate the well profitability. This study proposes the use of different machine learning techniques to predict the EUR as a function of the completion design including the lateral length, the number of stages, the total injected proppant and slurry volumes, and the maximum treating pressure measured during the fracturing operations. A data set of 200 well production data and completion designs was collected from oil production wells in the Niobrara shale formation. Artificial neural network (ANN) and random forest (RF) techniques were implemented to predict EUR from the completion design. The results showed a low accuracy of direct prediction of the EUR from the completion design. Hence, an intermediate step of estimating the initial well production rate (Q(i)) from the completion data was carried out, and then, the Q(i) and the completion design were used as input parameters to predict the EUR. The ANN and RF models accurately predicted the EUR from the completion design data and the estimated Q(i). The correlation coefficient (R) values between actual EUR and predicted EUR from the ANN model were 0.96 and 0.95 compared with 0.99 and 0.95 from the RF model for training and testing, respectively. A new correlation was developed based on the weight and biases from the optimized ANN model with an R value of 0.95. This study provides ML application with an empirical correlation to predict the EUR from the completion design parameters at an early time without the need for complex numerical simulation analysis. The developed models require only the initial flow rate along with the completion design to predict EUR with high certainty without the need for several months of production similar to the DCA models.