Recognition Of Flow Petterns And Research On Flow Characteristics Of The Horizontal Oil-water Two-phase Flow

Horizontal oil-water two-phase flow exists in oil exploration and engineering transportation commonly.Accompanied by delamination interface instability,oil water morphology fluctuations,the formation,growth,and transformation of flow patterns are extremely complex and the flow evolution behavior is characterized by low periodicity,irregularity,and variability.Therefore,adopting an accurate and effective two-phase flow patterns identification and characteristic analysis methods has significant scientific affects and potential engineering application value for understanding the complex flow structure and revealing its dynamic evolution mechanism.This paper performs systematic research work based on the time series analysis and image learning.The specific work is as follows:(1)Extracting the physical information function of oil-water two-phase flow based on the multivariate variational mode decomposition algorithm.In this paper,we introduce a multivariate variational mode decomposition(MVMD)algorithm from the perspective of time series decomposition,which is used to decompose and extract information from oil-water-conductivity fluctuation signals.The intrinsic mode functions after decomposing are used as intermediate information functions to calculate the multiscale multivariate fuzzy entropy(MMFE)to distinguish different horizontal oil-water flow patterns.Then,a multivariate empirical mode decomposition(MEMD)algorithm is selected for comparative verification.The experimental results indicate that the MVMD-MMFE algorithm can extract information features from fluid conductivity signals effectively,and reveal the complex structural changes of oil-water two-phase flow preliminarily.(2)Time-frequency characteristics analysis of oil-water two-phase flow based on the improved Hilbert-Huang transform.In order to analyze the dynamic time-frequency features,the method of multivariate variational mode decomposition is applied instead of the empirical mode decomposition in Hilbert Huang Transform(HHT),which can extend the dimension of input signals from one-dimensional to multi-dimensional time series.Similarly,The Hilbert transform and marginal spectrum are employed to describe the joint time frequency energy fluctuation of oil-water two-phase flow.The results show that the method can reveal the dynamic characteristics of nonlinear and non-stationary fluid signals from the perspective of time-frequency analysis,and identify four typical oil-water two-phase flow patterns effectively.(3)Achieving information mining and flow patterns classification of oil-water flow images based on the deep learning.we construct an oil-water image dataset under the actual production application scenarios.Then,applying the YOLOv5 neural network to learn three oil-water flow pattern images.Meanwhile,choosing the YOLOv4,SSD,and Faster-RCNN networks to perform flow pattern prediction analysis on the trained model.The experimental results show that YOLOv5 network has obvious advantages in flow patterns recognition accuracy,speed,and the size of the network model.Besides,it can meet actual production needs with excellent accuracy,rapidity,and lightweight characteristics.