Research on Fusion Algorithm of Machine Learning and Oilfield Fluid Property Estimation

In oilfield development, fluid properties (e.g., viscosity, density, and gas-oil ratio) are key parameters affecting recovery efficiency, reservoir management, and design of production enhancement programs. With the change of exploration research ideas, the innovation of geological understanding and the advancement of operation technology, the traditional fluid property estimation methods rely on experimental data and empirical formulas, which are difficult to maintain high accuracy in complex environments and have high costs. With the development of machine learning technology, data-driven methods show obvious advantages in dealing with nonlinear relationships and realizing fast prediction. There is a direct correlation between hydrocarbon component types and contents and reservoir fluid characteristics. In this paper, a fusion algorithm is proposed to combine traditional physical models with machine learning models to estimate oilfield fluid properties more efficiently. The proposed fusion algorithm is experimentally validated to perform well in multiple oilfield fluid property estimation tasks with better accuracy and robustness than a single model. This study provides a new technical idea for oilfield fluid property estimation, which is expected to improve the accuracy and real-time performance of fluid property prediction in practical applications, and help the intelligence and refinement of oil and gas field development.