Data-Driven Methodology for the Prediction of Fluid Flow in Ultrasonic Production Logging Data Processing

Abstract

A new method for the determination of oil and water flow rates in vertical upward oil-water two-phase pipe flows has been proposed. This method consists of an application of machine learning techniques on the probability density function (PDF) and the power spectral density (PSD) of the power spectrum output of an ultrasonic Doppler sensor in the pipe. The power spectrum characteristic parameters of the two-phase flow are first determined by the probability density function (PDF) method. Then, the transducer signal is preprocessed by distance correlation analysis (DCA), and independent features are extracted by principal component analysis (PCA). The extracted features are used as input to a least-squares fit, which gave the oil flow rates as output. In the same way, the transducer signal is also preprocessed by partial correlation analysis (PCA), and independent features were extracted using independent component analysis (ICA). The extracted features were used as inputs to multilayer back-propagation neural networks, which water cuts as output. The present method was used to calibrate an ultrasonic Doppler sensor to estimate the flow rates of both phases in oil–water flow in a vertical pipe of diameter 159 mm. Predictions of the present method were in good agreement with direct flow rate measurements. Compared to previously used methods of feature extraction from the ultrasonic Doppler power spectrum signals, the present method provides a theoretical basis for the interpretation of ultrasonic multiphase flow logging data. Ultrasonic multiphase flow logging has potential application value in the production profile logging and interpretation evaluation of production wells with low fluid production and high water cut.