Comparative Study on Supervised Learning Models for Productivity Forecasting of Shale Reservoirs Based on a Data-Driven Approach

Abstract

Due to the rapid development of shale gas, a system has been established that can utilize a considerable amount of data using the database system. As a result, many studies using various machine learning techniques were carried out to predict the productivity of shale gas reservoirs. In this study, a comprehensive analysis is performed for a machine learning method based on data-driven approaches that evaluates productivity for shale gas wells by using various parameters such as hydraulic fracturing and well completion in Eagle Ford shale gas field. Two techniques are used to improve the performance of the productivity prediction machine learning model developed in this study. First, the optimal input variables were selected by using the variables importance method (VIM). Second, cluster analysis was used to analyze the similarities in the datasets and recreate the machine learning models for each cluster to compare the training and test results. To predict productivity, we used random forest (RF), gradient boosting tree (GBM), and support vector machine (SVM) supervised learning models. Compared to other supervised learning models, RF, which is applied with the VIM, has the best prediction performance. The retraining model through cluster analysis has excellent predictive performance. The developed model and prediction workflow are considered useful for reservoir engineers planning of field development plan.

Application of Machine Learning Method of Data-Driven Deep Learning Model to Predict Well Production Rate in the Shale Gas Reservoirs

“…Mohaghegh [26] quantitatively assessed the impact of shale gas development by analyzing the correlation and pattern recognition between productivity and other factors such as rock mechanical properties, hydraulic fracturing data and reservoir properties. Exploratory data analysis (EDA) has been used to predict the cumulative production of a shale play using a variety of supervised learning techniques, such as regression analysis, support vector machines (SVM) and gradient boosting tree machines (GBM) [27]. In addition, a classification and regression tree (CART) has been performed using regression analysis to identify the importance of input data and to select useful input data for production prediction, to improve the predictive performance of the model [27][28][29][30][31].…”

A Novel Stacking Heterogeneous Ensemble Model with Hybrid Wrapper‐Based Feature Selection for Reservoir Productivity Predictions

“…Feature selection, which involves the selection of a feature subset from the original input data by removing irrelevant, redundant, or noisy features while maintaining model performance, generally requires the application of dimensionality reduction techniques [19]. Feature selection is an important processing stage in prediction models [20] because it eliminates the dimensionality problems in datasets with a large number of features [21], 2 Complexity helps ensure that soft-computing models are well trained, and reduces overfitting risks [15]. ere are three main feature selection techniques: filterbased, wrapper-based, and embedded [22,23].…”

“…Other research has also found that filter-based feature selection methods perform more poorly than wrapper-based feature selection methods [14]. Han et al [15] proposed a hybrid machine learning model which combines an individual cluster analysis (K-means clustering, partitioning around medoids clustering, and hierarchical clustering) and an individual regressor (random forest, Gradient boosting tree, and support vector machine) for productivity forecasting of shale reservoirs and validated the performance of their proposed model on the data set with 129 well logs. For the purpose of avoiding the overfitting problem and reducing computation time, a wrapperbased feature selection process was utilized, which also improves the model's interpretability.…”

“…However, the inherent shortcomings in the individual variable selection algorithm may lead to mistaken important feature identification results [18]. For example, Han et al [15] solely employed RF to select the main features, which may neglect the main features which may have a significant linear correlation with the forecast target; and the method utilized by Brantson et al [13]…”

Machine learning based decline curve analysis for short-term oil production forecast

“…The required datasets were obtained from 150 wells, targeting shale gas, stationed at Eagle Ford shale formations. Reservoir properties, well stimulation and completion were considered as key input parameters whilst the cumulative production of gas during a span of 3 years was identified as the target variable.Although (Aditya et al, 2017;Han et al, 2020) results were promising, the applicability of their methodology depends heavily on the presence of specific geological, well stimulation and completion data, and the quality and accuracy of the data have a big impact and influence, any anomaly in data consequently make their results less promising.In the context of deep learning, (Luo et al, 2019) built non-linear models using RF and Deep Neural Network (DNN) algorithms to forecast the cumulative production of oil during a span of 6 months. The whole dataset was obtained from around 3600 wells positioned at Eagle Ford formations.…”