Gas/Water Transient Rate Analysis and Parameters Evaluation by Integrating Pseudopressure-Based Approach with Deep-Learning Method

Abstract

Abstract Traditional transient rate analysis (RTA) based on single-phase flow equation has been limited to two-phase gas/water flow case. The primary complication associated with the adaptation of these single-phase solutions to wells exhibiting two-phase flow is the significant difference of gas/water PVT properties and the interaction between pressure and saturation during depletion process. As a result, the material-balance time based on single-phase assumption is not practical for analyzing two-phase flow from gas wells with gas/water flow characteristics. Meanwhile, there is still a lack of an automatic interpretation model on unknown parameters through history matching with measured production data, which increases RTA application threshold in the field of multiphase production data analysis. This work presents an integrated method of forward modeling of production performance and inverse interpretation of unknown parameters in the depletion-driven reservoirs. In the forward part, the gas/water-phase integral transformations including pseudopressure and beta factor are proposed to handle the two-phase nonlinearities in the governing flow equations of gas and water phases. The rescaled well-performance equations subject to variable operating conditions are presented through integrating the existing single-phase liquid formations with Duhamel convolution. The normalized gas/water-phase pseudopressure and material-balance pseudotime are introduced to meet the benchmark requirement of using single-phase liquid type curves. In the inverse part, an automatic interpretation model of gas/water RTA is established based on one-dimensional convolutional neural network (CNN). CNN-based proxy model is built through training large synthetic data set from forward model. Automatic inversion of the parameters is completed by minimizing the fitness function from the loss between proxy model and measured data with the help of optimization-based algorithm. The new method is validated against numerical simulation, covering a wide range of fluid properties and operation constraints.