A Rapid Multi-Objective Intelligent Decision Method for Shale Gas Fracturing Parameters Based on Machine Learning: A Case Study in the Changning Reservoir

Abstract

Abstract Proper design of fracturing parameters does not only increase the productivity and profitability of shale gas wells but also reduce the damage fracturing fluids cause to fluid-sensitive formations. Unfortunately, fracturing parameters design methods developed so far rely heavily on simulation runs and experience of field engineers. To lessen the computational burden and decrease the risks during fracturing, this study proposes a simple-yet-rigorous rapid intelligent multi-objective decision (RIMOD) method for shale gas fracturing parameters. The proposed RIMOD method includes two key sections, the first one is the squares support vector regression (LSSVR) model for objective evaluation and the other is non-dominated sorting genetic algorithm-II (NSGA-II) for finding the best shale gas fracturing parameters. Meanwhile, the flowback ratio of fracturing fluid (FBR) and first-month gas production (PROD) were considered as the objectives while fracturing parameters such as number of fractured sections, proppant amount and fractured length were selected as the decision variables. According to the data from sixty-six fractured horizontal wells from the Changning shale gas reservoirs, the intelligence decision process was completed in less than ten seconds. Furthermore, the accuracy of optimal scenarios in the final Pareto front was tested with the numerical model includes the geological features of the Changning reservoir. It was found that for both FBR and PROD the mean relative errors were less than 4%. It was concluded that, compared with simulation and empirical methods, the proposed RIMOD method could obtain fracturing parameters with greater efficiency and reliable precision. A field engineer could choose a scenario from the final optimal solution set based on the demand of the company. This could be reduction in formation damage or increased single well production. More importantly, once the data of a fractured well is obtained, this approach can be used to immediately establish an intelligent decision model for designing the fracture parameters of newly drilled wells.