Shale oil production prediction and fracturing optimization based on machine learning

With the advancement of horizontal well and hydraulic fracturing technology, the development of unconventional reservoirs such as shale oil has become a hot issue in the energy field. However, production tests have demonstrated that production is non-uniformly between stages due to reservoir heterogeneity and fracturing design limitations. Therefore, it makes sense to improve fracturing design to enhance shale oil development. In this paper, we propose a computational framework for shale oil production prediction and fracturing parameters optimization that couples machine learning and particle swarm optimization (PSO). Firstly, we construct a deep neural network (DNN) model database with 841 numerical simulation data as training set and validation set, and 87 field data as test set. Secondly, the hyperparameter optimized DNN are performed to predict the production performance. And the predictive performance compares to the random forest (RF) and support vector machine (SVM). Thirdly, coupled with DNN, PSO is performed to optimize fracturing parameters. Finally, conducting rapid fracturing design based on PSO optimization results and reservoir sweet spot distribution. The results reveal that DNN exhibit best production prediction accuracy compared to RF and SVM. The generalization ability of DNN is verified by accurate prediction performance of 4 cases with extreme parameters. Optimized fracturing parameters using PSO in an actual well resulted in 2969 m3 increases in cumulative oil and 68*104 USD increases in NPV. According to the optimization results of PSO in four extreme cases, the reasonable fracturing parameters for different reservoir quality are obtained, including the number of fracturing stages and clusters, the volume of single stage fracturing fluid and proppant. Our work guides engineers in rapid fracturing design while improving shale oil development effect.