Digital Twin and Simulation Studies Driven by Digital Transformation

_

This article, written by JPT Technology Editor Chris Carpenter, contains highlights of paper SPE 227319, “A Reservoir Dynamic Prediction Model Based on the REROSIM Method: Digital Twin and Simulation Studies Driven by Digital Transformation,” by Liangzhu Yan, SPE, Chengdu University of Technology; Zhiyuan Zhou, Yangtze University; and Bodong Li, Zhejiang Tensing Technology, et al. The paper has not been peer-reviewed.

_

Reservoir management in mature waterflooded fields demands accurate dynamic modeling and agile optimization under complex heterogeneity and ever-changing conditions. This paper presents a novel reservoir engineering reservoir simulation (REROSIM) approach—a data-driven interwell-connectivity model augmented as a digital twin—to predict reservoir dynamics and optimize operations in the Changqing oil field of China. The methodology represents the reservoir as a network of connection units linking injectors and producers, characterized by transmissibility and connected volume parameters instead of traditional gridblocks. History-matched with production and injection data by automated machine-learning calibration, the model captures time-varying well connectivity and flow paths with high speed and reasonable accuracy.

Introduction

Given the limitations of traditional modeling approaches, a clear need exists for a model that captures essential interwell connectivity physics with more fidelity than lumped capacitance-resistance models (CRMs), yet is significantly faster and more data-driven than full simulations. In recent years, data-driven interwell models have emerged to fill this gap. The presented approach treats the reservoir as a network of discrete conduits between wells, each with properties representing flow capacity and storage. By honoring material-balance and pressure-support principles, the approach can approximate waterflood performance, and its subsequent enhancements have improved its predictive power. Building an effective digital twin requires a model that is fast, automatically updatable, and sufficiently accurate to provide decision support, integrating field data continuously. Machine learning and artificial intelligence (AI) have been tapped to assist in achieving this goal—for instance, using neural networks to infer connectivity from production data—but purely data- driven models can lack physical interpretability.

In this context, a reservoir dynamic prediction model based on the REROSIM method is proposed by the authors which serves as the computational engine of a reservoir digital twin.

Theoretical Framework and Methodology

REROSIM and the Connection Unit Model.

At the core of the proposed approach is the concept of representing the reservoir as a network of interwell connections rather than a continuum of grid cells. This concept forms the backbone of REROSIM.

The REROSIM model simplifies a waterflood reservoir into a set of nodes (wells) connected by flow paths. Each injector/producer pair has an associated conductivity (transmissibility) and connectivity volume (effective pore volume) that govern fluid allocation and pressure support. Essentially, REROSIM solves a materially balanced flow distribution: it uses injection-rate signals and fluid material balance (similar to CRM’s mass-balance approach) to infer how much of each injector’s water goes to each producer (splitting coefficients), and how quickly (time-lag parameters). By fitting these to historical rates, the approach captures the interwell-connectivity matrix of the field. One can think of REROSIM as a coarse, meshless simulation on a graph, where each edge has properties analogous to a tube or channel connecting wells.