Objective Oil well production measurement is essential for assessing resource reserves and productivity, serving as a critical reference for resource development and management. The medium produced at the wellhead is typically a multiphase fluid. Current physical metering methods involve significant investment and complexity, while virtual metering requires numerous monitoring parameters, leading to implementation challenges and high computational costs.

Methods The produced fluid undergoes both temperature and pressure drops as it passes through nozzle throttling. To enhance the accuracy of the single-well virtual metering system and the convenience of maintenance, a new method for oil and gas metering was proposed based on nozzle throttling, leveraging the characteristics of differential pressure fluctuations and temperature difference signals. The mechanism equation for nozzle throttling flow, incorporating differential pressure, flow rate, and gas mass fraction, was derived. A deep neural network was established using nine characteristics as input parameters, including throttling temperature difference, mean pressure difference and standard deviation, to facilitate data-driven inverse prediction of gas mass fraction. Taking a 10 mm real nozzle as the test object, the experimental tests were carried out on a gas-liquid two-phase flow loop. In the tests, the converted velocity range of gas phase was 1.73–12.09 m/s while that of liquid phase was 0.03–0.35 m/s. The flow patterns tested included stratified flow, wave flow, slug flow, and annular flow.

Results The adaptive range of gas fraction was 0–100% for the mechanism equation, unaffected by changes in flow pattern, gas-liquid velocity, or system pressure. The errors in gas mass fraction and flow metering were within ±10%.

Conclusion The virtual metering of gas-liquid two-phase flow in oil nozzles based on physical-data fusion enables the virtual metering of gas and liquid flow using only the existing temperature and pressure measurement systems, without requiring information on storage, wellbore, or gathering and transportation pipeline network, nor additional measuring instruments. Additionally, low costs for data acquisition and modeling enhance the method's potential for widespread adoption and application.