Evaluation of Hydraulic Fracture Performance in the Viking Formation Using a Data-Driven Approach
Date
2020-09-14
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
ORCID
Type
Thesis
Degree Level
Masters
Abstract
The exploitation of oil from unconventional reservoirs has increased rapidly due to recent advances in horizontal drilling and multi-stage hydraulic fracturing. One of the main challenges in the exploitation of unconventional resources is the optimization of stimulation designs to maximize well performance. In recent years, data-driven methods have played a vital role in achieving operational, performance, and economic efficiencies in such resources.
In this study, data-driven predictive models were developed to evaluate hydraulic fracture stimulation design parameters and to identify oil production drivers in the Viking Formation. Both Multiple Linear Regression (MLR) and Random Forest (RF) models were used to analyze the data from 845 multi-stage hydraulically fractured horizontal wells in the Viking formation, Saskatchewan, Canada. Reservoir characteristics, such as well geographic location, average gas-oil ratio, net pay, and stimulation design parameters, including completion length, proppant concentration, proppant intensity, were included as inputs in the models. 365-day cumulative oil production (IP365) was used as a metric for well performance.
MLR and RF were successfully used to develop models for predicting IP365. The performance of both models was compared, and RF model was selected for further investigation due to its superior performance. The optimum combination of hyperparameters for RF was found using Random Search and Grid Search methods. Furthermore, the K-Fold Cross Validation approach was employed to validate the predictive and generalization capabilities of the model. The importance of reservoir and stimulation parameters with respect to well performance was determined using the Permutation Feature Importance technique.
Modeling results indicated that completion length has the largest impact on oil production, followed by proppant intensity (proppant volume by completion length) and reservoir characteristics such as net pay. Furthermore, a high gas-oil ratio was found to have a detrimental effect on oil production. Partial dependence plots were used to visualize the relationship between each predictor variable and the model output. Results showed that, on average, wells in the study area with lateral length greater than 750 m tend to have above-average IP365. Proppant intensities ranging from 0.37 t/m to 0.49 t/m seem to be effective. A diminishing effect on production is observed in wells treated with higher values of proppant intensity. Shapley additive explanations (SHAP) values were estimated to understand well performance on a well-by-well-basis. This method provided insights about the interactions between the predictor variables and the model output.
Results from this study will be helpful for stimulation and production engineers to optimize stimulation designs and maximize well productivity in the Viking Formation. This data-driven approach can be applied to any other plays and seamlessly integrated in workflow processes. Moreover, this methodology allows engineers and geoscientists to make informed decisions in rapidly changing environments.
Description
Keywords
Viking Formation, Machine Learning, Completion Design, Hydraulic Fracture, Well Performance
Citation
Degree
Master of Science (M.Sc.)
Department
Civil and Geological Engineering
Program
Civil Engineering