Optimization of field development scheme parameters using multivariant simulation and neural proxy-model

Abstract

This article presents an approach for determining the optimal parameters of a reservoir development system, based on a series of multivariate hydrodynamic simulations aimed at model adaptation and forecasting of technological indicators, incorporating neural network analysis. The rationale behind this algorithm is to enhance the accuracy and reliability of results during the early stages of design by simultaneously accounting for geological and hydrodynamic uncertain-ties. The software "tNavigator" was selected as the primary tool due to its extensive feature set tailored for this task. Using the Latin Hypercube algorithm, we conducted a multivariate adaptation of the initial hydrodynamic model. By analyzing the quality of the resulting model, we selected representative implementations for the baseline forecast. Based on the outcomes of baseline forecast and using the accumulated distribution function, we identified pessimistic, baseline, and optimistic scenarios for optimization calculations. These calculations were aimed at finding the most effective development system using the differential evolution algorithm.  

To ensure quality control and refine the optimal parameters obtained, we constructed a neural proxy model. According to the results of the study, we developed a procedure for obtaining desired estimates, which combines a wide range of uncertainties that define the variety of obtained solutions while also reducing the computational time required for simulations.