The prediction and optimization of Hydraulic fracturing by integrating the numerical simulation and the machine learning methods

Abstract

Hydraulic fracturing (HF) is an indispensable technique to economically develop unconventional oil/gas resources. More careful design and optimization of the field HF treatment can lead to a more satisfactory profit. Due to the uncertainty and complexity of the petroleum domain problems, the application of the machine learning (ML) method in such fields encountered many drawbacks in previous attempts. This paper proposed a new method, embodying the physical information, to find the most profitable completion design with high effectiveness and efficiency. The proposed model can accurately represent the Non-linear relationship between inputs and outputs. Meanwhile, incorporating the prior knowledge can enhance the generalization ability of the proxy model. The proposed model has been verified by two application scenarios. For both, the fracture propagation code and automated reservoir simulation code are integrated to generate the datasets. The input vector contains the treatment pressure of each stage, the treatment volume of each stage, and the cluster spacing. The output is the corresponding Net Present Value (NPV). Five proxy models are generated based on the proposed model, the regular neural network (NN), the Random Forest (RF), the Adaptive Boosting (AdaBoost), and the support vector machines (SVM) algorithms. By comparison, the prediction accuracy of the proposed model is superior to other models. The scaled Mean Squared Error (sMSE) of the predicted model increased by 16%, 346%, 266%, and 133%, respectively. Meanwhile, the proposed model has shown a higher generalization ability. The Relative Absolute Error (RAE) of the proposed model remains the lowest and the difference from other models becomes larger with the increase of the noise amounts. By introducing prior knowledge, the proposed model can predict the NPV more accurately and robustly.