Analysis of the Steam Pressure Prediction On Steamflood Process Using NARX Model

Shabira, Farah Judith Aida Nur (2023) Analysis of the Steam Pressure Prediction On Steamflood Process Using NARX Model. Other thesis, Institut Teknologi Sepuluh Nopember.

[thumbnail of 02111942000003-Undergraduate_thesis.pdf] Text
02111942000003-Undergraduate_thesis.pdf - Accepted Version
Restricted to Repository staff only until 1 September 2025.

Download (4MB) | Request a copy

Abstract

One of the methods to optimize oil production is Enhanced Oil Recovery (EOR) process using thermal injection methods. The most widely used method for thermal injection is the steamflood process. In steamflood method, steam pressure is one of the most important factors affecting the production of oil and gas, because the pressure is used to control the steam pipe choke area to control the steam source. This research aims to determine the effect of hyperparameter tuning on the accuracy of steam pressure prediction and the impact of the number of input lags on the steam pressure prediction. This study is also to know the robustness on the prediction result and the accuracy and error result of steam pressure prediction. The method uses Nonlinear Autoregressive with eXogenous inputs (NARX) model with Multi Layer Perceptron (MLP) regressor algorithm to predict the steam pressure. The prediction error and accuracy are evaluated using Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and relative accuracy metrics. This study is conducted in Jati Field, Sumatra, using sample data from three years and five months (1248 datasets). The exogenous input parameters used are the steam pressure and choke area from 13 areas, steam source, and total steam. The endogenous input parameter used is the steam pressure in the predicted area, with the prediction output is the steam pressure on the next day (t+1) and two days after (t+2). The model is tuned using default hyperparameters, along with activation function tuning using logistic and identity functions. The prediction result is observed using 5 fold k-fold cross-validation to see the model robustness and the performance. As a result, the logistic activation function tuning gained highest accuracy prediction for MAE metric with value as 0.036 with 96.24% accuracy. For RMSE metric, the highest accuracy is gained from identity activation function with value as 0.051 and 93.67% of accuracy. Furthermore, this study contributes to the field of petroleum engineering by presenting an effective methodology for accurately predicting steam pressure in the steamflood process. Also as providing a new and effective approach to predict steam pressure, which can be applied to other similar fields.
==================================================================================================================================
Salah satu metode yang dapat digunakan dalam mengoptimalkan produksi minyak adalah dengan proses Enhanced Oil Recovery (EOR) melalui metode injeksi termal. Pada injeksi termal, metode yang paling banyak digunakan adalah proses steamflood. Melalui proses ini, tekanan uap menjadi salah satu faktor utama yang memengaruhi produksi minyak dan gas, karena tekanan tersebut digunakan untuk mengontrol choke area pada pipa uap untuk mengontrol sumber uap. Penelitian ini memiliki tujuan untuk mengatahui pengaruh pengaturan hyperparameter terhadap akurasi prediksi tekanan uap dan pengaruhnya pada jumlah lag terhadap prediksi tekanan uap. Penelitian ini juga untuk melihat robustness, error, dan akurasi model pada hasil prediksi. Metode penelitian ini menggunakan model Nonlinear Autoregressive with eXogenous Inputs (NARX) dengan algoritma Multi Layer Perceptron (MLP) Regressor untuk mendapatkan prediksi tekanan uap. Nilai error dan akurasi dievaluasi menggunakan metrik Mean Absolute Error (MAE), Root mean Square Error (RMSE), dan akurasi relatif. Penelitian ini dilaksanakan di Lapangan Jati di Sumatera dengan menggunakan data sampel selama periode tiga tahun lima bulan yang terdiri atas 1248 dataset. Parameter input eksogen yang digunakan adalah tekanan uap dan choke area dari 13 area, sumber uap, dan total uap. Parameter input endogen yang digunakan adalah tekanan uap pada area yang diprediksi dengan output prediksi berupa tekanan pada satu hari setelah (t+1) dan dua hari setelah (t+2). Model diatur menggunakan default hyperparameter tuning serta activation function tuning meggunakan fungsi logistik dan identitas. Hasil prediksi ditelaah dengan k-fold cross validation sebanyak lima fold untuk melihat robustness dari model dan performanya. Hasilnya, activation function tuning logistik menghasilkan prediksi dengan akurasi tertinggi dengan nilai MAE sebesar 0,036 dengan akurasi sebesar 96,24%. Pada metrik RMSE, activation function tuning identitas memiliki nilai akurasi tertinggi sebesar 0,051 dengan akurasi sebesar 93,67%. Penelitian ini memberikan kontribusi pada bidang teknik perminyakan dengan menyajikan metodologi yang efektif untuk memprediksi tekanan uap secara akurat pada proses steamflood, sekaligus memberikan pendekatan baru dan efektif untuk memprediksi tekanan uap, yang dapat diterapkan pada bidang lain yang serupa.

Item Type: Thesis (Other)
Uncontrolled Keywords: NARX model, Enhanced Oil Recovery, Steamflood. Steam pressure; Model NARX, Enhanced Oil Recovery, Steamflood, Steam pressure
Subjects: Q Science > Q Science (General) > Q325.5 Machine learning. Support vector machines.
Q Science > QA Mathematics > QA336 Artificial Intelligence
T Technology > TN Mining engineering. Metallurgy > TN879.5 Petroleum pipelines
Divisions: Faculty of Industrial Technology and Systems Engineering (INDSYS) > Mechanical Engineering > 21201-(S1) Undergraduate Thesis
Depositing User: Farah Judith Aida Nur Shabira
Date Deposited: 05 Oct 2023 08:58
Last Modified: 05 Oct 2023 08:58
URI: http://repository.its.ac.id/id/eprint/104172

Actions (login required)

View Item View Item