Sand Particle Monitoring for the High-Production Gas Well Based on EMD-CNN Method

Sand-carrying annular flow is a commonly observed flow pattern in high-production gas wells, and accurate real-time monitoring of sand particle behavior within the annular flow at the wellhead is essential for efficient industrial production. This study aims to establish a method for identifying particle migration behavior and size in the wellbore's annular flow, utilizing empirical modal decomposition (EMD), statistical analysis, Hilbert-Huang transform (HHT), and convolutional neural network (CNN). Through time and frequency analysis, the study successfully elucidates sand migration behavior, including sand carrying by the gas core (IMF 1) and by the liquid film (IMF 2, IMF 3), and verifies these behaviors. ResNet-50 was determined as the best CNN model for particle size identification, and further optimization of its structure improved the accuracy of particle size recognition by 12% to 83.7%. These findings provide a novel approach to the study of solid phase particle detection in multiphase flow, and may significantly enhance recognition accuracy through model optimization. This methodology offers valuable support for the development of intelligent oilfields.