Optimization of Fracturing Parameters with Machine-Learning and Evolutionary Algorithm Methods

Oil production from tight oil reservoirs has become economically feasible because of the combination horizontal drilling and multistage hydraulic fracturing. Optimal fracture design plays a critical role in successful economical reservoir. However, many complex parameters such as spacing half-length make fracturing treatments costly uncertain. To improve design, it is essential to determine reasonable ranges for these evaluate their effects on well performance economic feasibility. In traditional analytical numerical simulation methods, simplifications assumptions are introduced artificial characterization gas percolation mechanisms, implementation process remains complicated computationally inefficient. Most previous studies big data-driven parameter optimization have been based only single output, expected ultimate recovery, few integrated machine learning with evolutionary algorithms optimize time-series prediction objectives. This study proposed novel approach, combining model parameters. We established significant number static dynamic data sets representing geological developmental characteristics simulation. Four production-prediction models machine-learning methods—support vector machine, gradient-boosted decision tree, random forest, multilayer perception—were constructed mapping functions between properties production. Then, parameters, best machine-learning-based predictive was coupled four algorithms—genetic algorithm, differential evolution simulated annealing particle swarm optimization—to investigate highest net present value (NPV). The results show that among established, perception (MLP) performance. Among algorithms, (PSO) not fastest convergence speed but also value. optimal area were identified. hybrid MLP-PSO represents robust convenient method forecast by reducing manual tuning.