DISCHARGED-WATER TEMPERATURE PREDICTION IN UPSTREAM OIL SURFACE FACILITY USING NONLINEAR AUTOREGRESSIVE NEURAL NETWORK MODEL WITH EXOGENOUS FEATURES

• Main Author: Partogi Hasudungan Silaen, Iwan
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/73257
• Institution: Institut Teknologi Bandung

<p align="justify">Discharged-water from the upstream oil industry is produced by surface facility that processes fluid resulted from drilling in which the characteristic is set by the Regulation of the Ministry of State for Environment Number 19 Year 2010, which states that the discharged-water temperature resulted from the upstream oil industry needs to be in between 40 – 45 degrees celsius. To keep the water temperature at this range, the produced water from the well is usually decreased. The decrease in production cannot be done arbitrarily because it will affect the oil production. Therefore, a model that can predict the discharged-water temperature based on the produced water from the well is needed. In this final project, a nonlinear autoregressive artificial neural network with exogenous features is utilized to handle the stated problem. This model is chosen because of its capability to handle complex, nonlinear, and temporally dependent problem, just like the interaction which occurs in surface facility. The data used in modeling comes from PT Pertamina Hulu Rokan in which there are several missing data, therefore imputation is done on the data using 2 methods, namely linear interpolation and random forest. Models for the simulation consist of normal artificial neural network and nonlinear autoregressive artificial neural network with exogenous features. Each model does prediction with data that has been imputed with linear interpolation and random forest, and later the performance is compared with naive forecasting method and real unimputed data. The simulation results show that nonlinear autoregressive artifical neural network with exogenous features with random forest imputation training data performs the best. This model is hoped to assist the upstream oil industry to determine the produced water limit that fulfills government regulation as well as resulting the smallest profit loss possible.