A Novel Prior-Guided Hybrid VFM Framework Integrates Mechanistic and Data-Driven Approaches in Oilfields

Abstract Real-time assessment of oil well production stands as a pivotal factor in the effective management and optimization of oilfield production and the gathering pipeline networks. Precise and timely real-time quantification of oil, gas, and water phases is of paramount importance, serving as a compass for on-site personnel in steering decision-making processes. Developing methods to accurately measure the flow rates of individual wells without expensive hardware is crucial for reducing production costs in the oil field industry. The burgeoning appeal of virtual flow metering (VFM), leveraging on-site sensor data encompassing pressure and temperature, has surged within oilfield production systems. In this paper, we propose a novel Prior-Guided Hybrid VFM framework, engineered to glean a richer tapestry of information from signal data. Initially, we introduce a Prior-Guided VFM Module, assimilating the foundational Electric Submersible Pump (ESP) model as prior knowledge in the data-driven model. This integration of ESP model guides the model training purposefully and simultaneously adjusts coefficients within the ESP model during model learning. This module amalgamates mechanistic models with data-driven models to yield enhanced precision in the outcomes obtained. Subsequently, we formulated a Secondary Hybrid Training Module integrating self-attention mechanisms. This module aims to extract potential relationships of multidimensional data in a high-dimensional space and fuses the results of the mechanism-containing module to ultimately achieve hybrid prediction. The empirical validation of this framework is conducted using actual data derived from an offshore oilfield. Comparative analysis against five data-driven models substantiates the efficacy of the proposed framework. Results demonstrate that this approach not only performs better but is also suitable for few-shot scene. This work provides profound insights into the integration of mechanistic and data-driven approaches in VFM. The significance of this study lies in reducing the costly hardware and manpower expenses in oilfield development, which is crucial for the advancement of smart oilfields.