MEMPREDIKSI LAJU PRODUKSI MINYAK, GAS, DAN AIR MENGGUNAKAN RECURRENT NEURAL NETWORKS : STUDI KOMPARATIF DENGAN LSTM, GRU, DAN VANILLA RNN

• Main Author: Putri Aurelia, Salsabila
• Format: Final Project
• Language: Indonesia
• Subjects: Pertambangan dan operasi berkaitan
• Online Access: https://digilib.itb.ac.id/gdl/view/73481
• Institution: Institut Teknologi Bandung

This research focuses on using machine learning techniques to predict well performance in the petroleum industry. Traditional methods like Decline Curve Analysis and reservoir simulation have limitations in accurately predicting production performance. Therefore, the study explores the use of machine learning algorithms, specifically LSTM, GRU, and Vanilla RNN, to forecast well performance production over time. The methodology involves collecting and preprocessing data, then splitting it into training, development, and test sets. The input data is scaled using min-max scaling. The models are implemented with specific architectures and trained using the Adam optimization algorithm and mean squared error as the loss function. The training process includes 50 epochs and early stopping to prevent overfitting. Hyperparameter tuning is conducted to optimize model performance. The results indicate that the LSTM model outperforms the GRU and Vanilla RNN models in terms of learning and generalization performance. Sensitivity analysis suggests that the GRU and Vanilla RNN models may not be suitable for the given task or dataset. The optimal hyperparameter values for the LSTM model are determined through sensitivity analysis. In conclusion, the LSTM model with optimized hyperparameters demonstrates the best performance for forecasting flow rates in well-performance prediction. This study provides insights into the effectiveness of different RNN architectures and contributes to predictive modeling in the oil and gas industry.