Machine learning for optimal ultra-fine cement plugging system in simulated high permeability sandstone reservoirs

Abstract

The properties of a cement system dictate its potential applications, yet creating a reliable cement plugging system with controllable setting times, robust injection capacity, and high compressive strength often requires a lot of time and resources in advanced oilfield development. To address this issue, a machine learning-based XGBoost model was created to optimize plugging system performance, reduce costs, and enhance efficiency by analysing the initial setting time, final setting time, viscosity, and compressive strength of ultra-fine cement. The results demonstrate an impressive accuracy of 96.08%, 96.10%, 94.35%, and 90.42% for these respective parameters, with a prediction time of 0.0068 seconds. In response to sandstone reservoirs with permeability greater than 500 mD, an intelligent optimization integrated model was established, resulting in three plugging system formulations that meet the requirements. The injection speed was 1 mL/min and the slug combination used was 0.2PV polymer and 0.7PV plugging system. The maximum injection pressure change recorded remained below 0.8 MPa. The plugging rate achieved was over 90%, reaching a maximum of 96.28%. This study provides theoretical guidance for the construction and injection process in the mining field test area and also offers valuable insights for the intelligent development of oilfields.