Perpustakaan Digital ITB

X Field is a mature oil field that has been producing since 1972 and now requires extensive workover activities to maintain and improve production efficiency. Although a significant volume of hydrocarbons remains unrecovered, declining well performance makes it increasingly important to identify workover candidates in a timely and systematic manner. This study proposes a data-driven framework to support workover planning using limited historical production data. The proposed framework integrates production forecasting and supervised classification into a single workflow. Long Short-Term Memory (LSTM) models are employed to forecast future oil and water production. The forecasted outputs are then used as inputs for a supervised classification model to determine the required workover type. The study focuses on five representative wells (XX003, XX004, XX015, XX016, and XX017) which are selected based on data quality, data quantity, and monthly average production. Meanwhile, there are four types of workover job incorporated, including re-perforation, revised packer, water shut off, and open new zone which will divided into Water shut off (WSO) and Perforation job. These jobs are selected based on the effectiveness to increase oil production rate and its operational frequency. This paper is divided into three main parts. First, the data is prepared by cleaning and selecting the most relevant features. Second, an LSTM-based model is developed to predict future oil and water production. Finally, a classification model is built and combined with the forecasting model to support decision making. Data involved encompassing monthly production data, workover history, and reservoir properties which proceed to create a unified and standardized data set for further analysis. In addition, the feature selection results indicate that future oil production is predicted using the most recent oil rate from well tests together with porosity, while water production is predicted to be using only the latest water rate from well tests. For the classification task, the model uses the predicted oil and water rates, as well as the most recent oil and water gradients to classify the data into “WSO”, “Perforation” and “NO” labels. The forecasting models are evaluated using MAPE, MAE, and RMSE metrics. A detailed error analysis is also conducted to distinguish error patterns between high and low target values. The results show that the forecasting model provides reliable predictions for up to three months ahead. For wells with high production rates, the average prediction error ranges from 0.1% to 9% for oil and 0.1% to 7% for water. For moderate to low-rate wells, the deviations are approximately 20 STBD for oil and between 138 and 300 STBD for water. The classification model was developed using a wrapper approach, in which multiple algorithms were evaluated. Based on the performance metrics—accuracy, precision, recall, and F-score. The Extra Trees Classifier achieved the best results, with an overall accuracy of up to 80.3%. on both the training and validation datasets. It successfully predicts the required workover type for wells XX003, XX004, XX015, and XX016, while it shows limitations in identifying water shut-off requirements for well XX017 when using forecasted inputs because the data availability to train “WSO” class is much lower than the other class. This means with larger number of sample and training data, it can robust the model’s performance. Overall, the proposed framework demonstrates strong potential as a practical and efficient decision-support tool for optimizing workover planning in mature oil fields, while reducing human intervention and maintaining consistent decision quality.