Assessing the remaining hydrocarbon potential in highly deviated wells in Malaysia Field is challenging because it is difficult to deploy logging operations because of the wellbore angle. To address this challenge, a machine learning (ML) model is adopted to predict behind-casing opportunity (BCO) results using field data. The ML model is integrated into the Production Enhancement Candidate Generation and Screening (PECGS) artificial intelligence (AI) workflow, which aims to identify new candidates for enhancing production. High-quality petrophysical properties and production history data are essential input features to implement this methodology. Before executing the ML model, numerous efforts to perform feature engineering on the data are needed to ensure the high accuracy of the BCO prediction. Understanding the difficulty level of these work steps reveals that extracting the required features from petrophysical data for multiple fields involves many tedious manual steps. These steps are time-consuming and can lead to errors with repetitive and monotonous tasks for hundreds of wells. An automated system is developed to minimize manual human intervention and improve data accuracy.

This paper presents a novel data science approach to accelerate the petrophysical data feature engineering process. This involves automating the petrophysical data preprocessing and quality control (QC) steps. Such automation offers high-quality well data and reservoir properties such as well trajectory, average porosity, net pay thickness, and the permeability thickness (Kh) used to predict BCO. An automated system incorporating comprehensive data analytics and quantile regression forest (QRF) techniques was developed. The system carries out the well data inventory, trajectory verification, formation evaluation result consistency check, and auto compilation of transformed petrophysical features. Once data analytics is incorporated into the system, within a few minutes it can output comprehensive well inventory reports that verify the well trajectories for a massive number of wells in a field. The QRF algorithm also helps rank the consistency and uncertainty of formation evaluation results across the field, which eventually guides the petrophysicist in determining the best result for reservoir properties summary computation. Additional validation process benefits enable missed pay detection, providing petrophysical results in zones with previous data gaps. In the final stage, the system also automatically sorts out a structured output that combines properties for multiple production zones to facilitate the usage across the ML BCO workflow.

This automated system offers better efficiency in feature extraction and more accurate answers in BCO prediction. One of the experiment fields has successfully increased the registered BCO candidates in highly deviated wells with the improved reservoir properties availability that result from the automated system. The solutions were applied to 24 additional fields in the Malaysia region. An estimated time savings of 500 hours was achieved in generating the highest quality reservoir properties ready to be used in ML BCO. The saved hours also translated into energy savings in computing power and drive toward a net-zero carbon footprint acceleration with a reduction of ~75 kg of CO2 emissions.

You can access this article if you purchase or spend a download.