Creation of a proxy model for the selection of optimal parameters of the development system using horizontal wells with multistage hydraulic fracturing

Abstract

The current state of oil production is characterized by a decline in the proportion of reserves located in conventional reservoirs. The development of complex and highly complex reservoirs is becoming increasingly important. A significant proportion of oil reserves are concentrated in low-permeability reservoirs that are no longer economically viable to develop using traditional field methods. The aim of the work is to create a predictive tool for the selection of optimal parameters of the development system using horizontal wells and multistage hydraulic fracturing of reservoirs. To solve this problem, a fully connected neural network model was trained that predicts the parameters of the production profile as a function of the initial geological and physical conditions and the parameters of the development system. The architecture of the resulting neural network includes 3 linear layers of 300 neurons each. The training sample for the model was the results of multivariate calculations on a synthetic hydrodynamic model simulating the operation of a development element using a horizontal well with multistage hydraulic fracturing in depletion mode. The developed model can be useful in solving the problem of designing a development system in new or undrilled areas of low-permeability oil fields.