Choke-Based Production Calculation and Size Optimization for Shale Oil during the Spontaneous Flow Period

Summary Shale oil wells exhibit rapid production decline and pressure instability during the spontaneous flow period, characterized by typical unsteady-state flow behavior. The currently widely used decline curve analysis method and most machine learning prediction models are essentially data-driven or empirical statistical modeling methods for producing dynamic responses. Prediction is achieved mainly by establishing a mapping relationship between production data or establishing a training set of reservoir parameters and production dynamic data. These methods do not consider physical flow processes. In addition, as a direct control factor for both production and pressure, the choke has received limited attention in production modeling research. In fact, immediate changes following choke size adjustments often involve significant production and pressure fluctuations, which current mainstream methods fail to accurately capture, often leading to computational inaccuracies. Consequently, we have developed a methodology for calculating production rates during the spontaneous flow period in shale oil reservoirs. By comprehensively considering production rates, pressure decline magnitudes, and choke change frequencies during the spontaneous flow period, the entire spontaneous flow period is segmented into multiple substages. Within each substage, a nodal analysis approach is used to integrate three flow processes: reservoir, wellbore, and choke. Through iterative calculations across stages, the production dynamic throughout the entire spontaneous flow period is subsequently calculated. Based on this, considering the direct impact of the choke size on the production, we further constructed an optimization model for the choke size. This model employs cumulative oil production as the objective function and utilizes genetic algorithms (GAs) as the optimization methodology. Subsequent research has shown us that the optimization efficiency of a pure GA is relatively low. Therefore, we integrated random forest (RF) algorithms as surrogate models, ultimately creating a surrogate-assisted genetic algorithm (SAGA) that combines GAs with RF techniques. We applied the aforementioned methodology to conduct production calculations for actual shale reservoir fractured horizontal wells and compared the results with monitoring data. The findings show that the production, pressure, and water-cut metrics from this method have an overall accuracy exceeding 85% when matched with historical dynamic data. Meanwhile, the improved SAGA has increased the optimization speed by more than 70% compared with GA, and the quality of the optimal solution is very close. In the optimization of two real fractured horizontal wells, SAGA achieved an increase in cumulative oil production in both cases.