Prediction of gas production potential based on machine learning in shale gas field: a case study

作者：Shuo Zhai, Shaoyang Geng, Chengyong Li, Yufeng Gong, Min Jing, Yao Li · 发表于：Energy Sources Part A Recovery Utilization and Environmental Effects · 年份：2022 · DOI：10.1080/15567036.2022.2100521 · 被引用次数：21 · 研究领域：Hydraulic Fracturing and Reservoir Analysis、Hydrocarbon exploration and reservoir analysis、Reservoir Engineering and Simulation Methods

Productivity prediction is an important aspect of oil and gas exploration and development. As the amount of field data has increased, traditional engineering methods have begun to face challenges. The prediction of shale gas well production is influenced by geological, drilling, and completion characteristics. The relationship between variables is highly nonlinear and non-intuitive. In this study, 384 production well data were collected, including 14 input features and one output feature, with a total of 5,760 data points, 80% of which were used for training processes and 20% for testing processes. Four machine learning methods, namely an extreme gradient lifting tree, random forest, artificial neural network and support vector machine, were used to construct a shale gas well productivity prediction model. It was found that the machine learning method could accurately can well reflect the nonlinear relationship between the influencing factors and productivity. After the data were entered into the model training, regression evaluation indexes, such as the determination coefficient R2, were used for measurement. Compared with other models, the support vector machine regression model with gamma (kernel coefficient) and C (punish coefficient) values of 1 and kernel function as the radial basis function have smaller errors. The mean absolute, mean square, and root mean square errors were 0.062, 0.006, and 0.077, respectively. The R2 values for the training and testing sets were 0....