Flow Pattern Identification And Pressure Drop Calculation On Three-Phase Flow In A Horizontal Pipeline

The transportation of oil, gas and water is the important point in oil industry development and gathering transportation project. Multi-phase transportation technology has the characteristics of reducing pipeline investment and working cost and so on, so it is increasingly used in gathering pipeline nets and oil pipeline transportation nets in oil fields,in which flow pattern identification is the means while the pressure drop calculation is the goal. Therefore, it is of great significance to perform an investigation on oil/gas/water three-phase flow in pipelines.The experiment of three-phase flow was carried out on the large multiphase flow loop at China University of Petroleum, in which HVIW150 base oil, softened water and air were used as experimental media. Three main flow patterns, including wavy stratified flow, slug flow and semi-annular flow, were observed during the experiment,which accord with the flow pattern figure of Lee. Transmitters of pressure and differential pressure were installed on the loop to collect the flow signals including pressure and differential pressure. The denoise method is the maximum modular value of wavelet. By programming in Matlab software, the collected signals and simulated signals were denoised successfully. Based on the denoised data of differential pressure in different flow pattern, the characteristic vectors were calculated including fractal characteristic vectors, statistical characteristic vectors and so on. And the fractal characteristic vectors include the correlation dimension, the maximum Lyapunov index and the Hurst index which were calculated by G-P algorithm, Wolf algorithm and R/S analysis method respectively. According to the results, the flow characteristics were analyzed that the three fractal characteristic vectors could be used as the bases for flow pattern identification. Statistical characteristic vectors include multi-scale information entropy, PSD characteristics and PDF characteristics and so on, and they are good matches to the fractal characteristic vectors.Twelve calculated characteristic vectors were input into the BP Neural Net, and the layer number of the net, the note number of the hidden layer, the learning rate and the excitation function were obtained. By training it, the veracity is 76.7%. Then the Genetic Algorithms was combined into the Neural Net, and the veracity is 90%.Sampling analyses were made for oil/water mixture with sampling probe, and for wavy stratified flow, relatively obvious gradient were found. Then the Gas-W/O-O/W model was selected, and the extended velocity was used, too. And the extended velocities under different oil and water heights were calculated, so that the new pressure drop model of wavy stratified flow was established. And the result accords with the experiment values.