VIRTUAL MULTIPHASE FLOW METER USING DYNAMIC NEURAL NETWORK BASED ON MEASUREMENT DATA OF ORIFICE, CHOKE VALVE, AND WELLHEAD

• Main Author: Anom, Fanzi
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/74335
• Institution: Institut Teknologi Bandung

Virtual Multiphase Flow meter (VMPFM) is intended to replace Multiphase Flow Meter (MPFM), an equipment to measure flow rate of gas, oil, and water in real time which does not require fully separation. VMPFM can be used as back-up measurement during well testing of oil and gas when MPFM is unavailable as well as comparison of MPFM. In this study, VMPFM modeling was carried out using four methods of artificial neural networks (ANN), including: Shallow Feed Forward Back Propagation (SFFBP-ANN), Deep Feed Forward Back Propagation (DFFBP-ANN), time series time delay (TSTD-ANN), and nonlinear autoregressive network (NARX-ANN). Several variations during training stage were made to see the accuracy performance of each model, including variations in the number of hidden layers, the number of neurons, the learning algorithms, and the time sample delay. Nine real data parameters during well testing in offshore platforms, from May 2020 to Oct 2021, are used for VMPFM modeling including parameters of wellhead pressure and temperature, orifice flow meter parameters, choke valve opening parameters, and multiphase measurement results by MPFM. The data set is divided into two parts where the first data set is used for training the model and the second data set is used for validating the model. The simulation results show that the TSTD-ANN method with a sample delay time of 30 and the addition of the amount of training dataset gives the most accurate measurement results and fast training duration (less than 7 minutes). The best average discrepancy value of flow rate between VMPFM and MPFM measurements during the validation stage is 6.0% of the gas flow rate, -16.4% of the oil flow rate and -2.4% of the water flow rate.