Eventos Anais de eventos
ENCIT 2022
19th Brazilian Congress of Thermal Sciences and Engineering
SEVERE SLUGGING IDENTIFICATION USING LONG-SHORT TERM MEMORY NETWORKS
Submission Author:
Carlos Mauricio Ruiz Diaz , SP , Brazil
Co-Authors:
Carlos Mauricio Ruiz Diaz, MARLON MAURICIO HERNANDEZ CELY, Oscar Mauricio Hernandez Rodriguez
Presenter: Carlos Mauricio Ruiz Diaz
doi://10.26678/ABCM.ENCIT2022.CIT22-0324
Abstract
The production of oil and gas in offshore platforms involves a great number of potentially dangerous events of great complexity. Severe slugging is a flow anomaly that may result in flooding of production facilities and decrease in productivity, with characteristics of transient pressures, volumetric fractions, and superficial velocities. A study will be carried out in a vertical pipe with two-phase flow of liquid and gas, aiming to simulate severe slugging. Liquid injection will be carried out by means of a positive displacement pump with a frequency inverter, and dense gas is provided to the system by a booster specially designed to work with gas. The gas injection will be periodically modified by flow control valves installed upstream the test line. Directional control valves are installed in each injection line together with pressure, temperature and flow sensors, whose measurements are stored in a computerized control system, where the operation of all the experimental equipment is monitored and the data is saved. The information collected was optimized with data science techniques and used to develop a predictive model based on the technique of long- and short-term memory networks to predict the types of severe slugging inside pipelines. A correlation matrix was developed to determine the relationship of parameters in the formation of severe slugging, including the specific density of the gas-liquid mixture, gas-liquid ratio (GLR), superficial gas velocity, superficial liquid velocity, gas volume fraction and liquid volume fraction. The parameters used to determine the performance of the AI model were the root mean square error (MSE), the Root Mean Square Error (RMSE) and the coefficient of determination (R²), for which values of 0.33%, 5.7% and 0.9983 were obtained, respectively.
Keywords
severe slugging, Long-Short Term Memory, Two-phase Flow, dense gas
