Petrovietnam Journal

AN ARTIFICIAL NEURAL NETWORK APPROACH TO OPTIMIZE THE WATER FLOODING OF BACH HO OILFIELD, OFFSHORE VIETNAM

2025-11-25T19:07:52+00:00

The predominant oil production offshore Vietnam originates from the Bach Ho basement reservoir, where the flow regime is highly complicated due to the heterogeneous spatial distribution of petrophysical properties such as porosity, permeability, and water saturation.
Therefore, the conventional reservoir-simulation-based methods for forecasting oil production often yield limited accuracy or require substantial time and effort to optimize the dynamic parameters. Recent advances in machine learning (ML) algorithms enable more rapid and accurate prediction of oil production rates from water injection rates at individual injection wells. Once the oil rate prediction is achieved using ML approach, the waterflooding optimization can then be achieved by any suitable optimization algorithm.
In this research, an artificial neural network (ANN) algorithm is used, yielding very good results: the correlation coefficients between the predicted and actual values are 0.98 and 0.95 for training and testing datasets, respectively. Subsequently, the Gauss-Newton optimization algorithm is applied to determine the optimal water injection rates for each injection well, aiming to enhance oil productivity. The results show that the newly optimized injection schemes yield an average oil production increase of 1.5%.

DEVELOPMENT OF A PRODUCTION ALLOCATION MODEL FOR TIE-IN OIL FIELDS AT VIETSOVPETRO ON THE CLOUD COMPUTING PLATFORM

2025-11-25T19:07:25+00:00

In the process of integrating oil and gas gathering and transportation between adjacent fields, production allocation plays a crucial role for tie-in oil and gas fields in ensuring the interests of investors. In recent years, this issue has emerged for tie-in fields on the continental shelf of Vietnam.
The production allocation model integrates multiple input data sources, including flow parameters and fluid properties, to process the outcomes. Experimental models are applied to determine phase state variations of stage-separated products under different temperature and pressure conditions within the gathering and transportation system, from the production wellhead to the final storage point.
This paper systematically presents the computational process for production allocation, the digital transformation of workflows, and algorithmic simulation to develop the AIT production allocation model on a cloud computing platform. The implementation of this model is expected to optimize production operations, enhance accuracy, and improve efficiency in production allocation for tie-in fields.

AI-INTEGRATED DOMAIN-SPECIFIC DATA MANAGEMENT: EXPERIENCE FROM DEVELOPING WELL-LOG DATABASE MANAGEMENT SOFTWARE IN THE CUU LONG BASIN

2025-11-25T19:06:59+00:00

The rapid advancement of artificial intelligence (AI), particularly Large Language Models, has significantly transformed data management and processing practices for both structured and unstructured data. This paper presents the experience of designing, implementing, and operating an AI-integrated geophysical well log database software for the Cuu Long basin at the Vietnam Petroleum Institute (VPI). The software was developed by leveraging AI to optimize the management, retrieval, and analysis of well logging data while ensuring strict security requirements.
The research results show that the product can be integrated with other supporting tools, contributing to an enhanced user experience.
Based on these findings, future development directions are proposed to optimize the system's efficiency and security.

APPLICATION OF PYTHON AND MACHINE LEARNING IN PROCESSING AND CLASSIFYING IMPORT-EXPORT DATA

2025-11-25T19:06:27+00:00

Vietnam’s import-export data is increasing substantially in both scale and complexity, creating significant challenges in standardizing and classifying customs declaration information. This study proposes an automated data-processing pipeline implemented in the Python programming language, with the objective of enhancing efficiency and ensuring greater consistency in the analysis of customs information. The input dataset comprises more than 10,000 real-world import-export records, which are processed through a structured sequence of technical steps, including product name normalization, unit conversion, computation of quantitative indicators, and keyword-based product group labeling.
The experimental results demonstrate that this processing pipeline operates effectively on medium-scale, high-complexity datasets, while considerably improving classification accuracy and ensuring uniformity across product categories. Based on these findings, the authors propose integrating machine learning models as a supplementary tool to enhance generalization capabilities and adaptability to exceptional cases - particularly relevant in a trade environment where product names are increasingly diverse, unstandardized, and continuously evolving.

AI APPLICATIONS IN MARKET INFORMATION COLLECTION AND MACROECONOMIC INDICATOR ANALYSIS

2025-11-26T19:06:22+00:00

Data analysis in economics is transitioning from traditional methods to AI-driven approaches, enabling automation in data collection and processing, as well as improving the ability to forecast market fluctuations, providing a foundation for more accurate business decision- making. This article presents the application of AI at the Vietnam Petroleum Institute (VPI) to develop a digital analytics solution for collecting market information, aimed at improving the efficiency of data integration and automated processing, forecasting economic indicators, and providing a virtual assistant to support interactive market information queries.

RESEARCH AND MANUFACTURE OF PORTABLE DEVICE TO DETECT PIPELINE DEFECTS USING THE MAGNETIC LEAKAGE FLUX METHOD

2025-11-26T19:05:08+00:00

To meet the needs of periodic health inspections of interconnected pipeline systems in factories such as refineries and ammonia plants, the research team developed a portable defect inspection device using the magnetic flux leakage (MFL) method. This device not only enables rapid inspection, but also features color imaging capability through a two-dimensional sensor array, 2D scanning mechanism, colorization technology and integrated mapping - features not previously available worldwide. The device is equipped with 64 flat Hall sensors arranged on a 2D array mounted on a curved base that conforms to the pipe’s radius. The sensor array and curved base are installed on a wheeled frame integrated with the magnetization system. The magnetization system consists of 3 blocks connected by a sliding plate mechanism, allowing adjustment to tightly fit pipes of different diameters. During operation, the device is manually pushed along the pipeline.
To enhance image resolution, the entire sensor is shifted by 1 mm along both the longitudinal and circumferential directions of the pipe after each measurement (image capture). The captured images are then integrated to produce a final image with 1 mm resolution.

DEVELOPMENT OF A WALL-CLIMBING ROBOT FOR CORROSION DETECTION OF VERTICAL OIL TANKS

2025-11-26T19:05:51+00:00

This paper presents the research, design, and fabrication of a prototype robot for corrosion detection on vertical cylindrical oil tanks using the magnetic flux leakage (MFL) inspection technique. The robot's overall design, propulsion system, electrical components, control architecture, and MFL inspection unit were meticulously optimized using specialized software, in compliance with IEC explosion safety standards. The completed prototype has a total mass of 39 kg and operates within a speed range of 10 - 50 mm/s.
The prototype then underwent testing on a vertical steel plate fabricated to simulate the physical properties of actual gasoline and oil storage tanks. Preliminary test results demonstrate the robot’s adherence to key performance criteria, including strong adhesion, efficient mobility on vertical surfaces, and high sensitivity to micro-defects. Specifically, the robot can traverse weld seams up to 10 mm in height, maintain stable movement at a maximum speed of 50 mm/s, and detect defects as small as 2 mm in depth on steel plates up to 14 mm thick.

INTEGRATION OF OPTICAL AND RADAR REMOTE SENSING IMAGES FOT CLASSIFYING AND MONITORING OIL SPILLS AT SEA

2025-11-26T19:05:30+00:00

Remote sensing data has been widely used in the world in the study of oil spill pollution at sea. This paper presents a study combining the use of Sentinel 2 MSI optical remote sensing and Sentinel 1 radar images to detect and classify oil spills. Sentinel 2 MSI data is used to calculate the OSI (oil spill index) based on visible bands, while Sentinel 1 data is used to calculate the backscatter value, from which oil spills are classified by the thresholding method. The integration of multi-type remote sensing data allows to enhance the density of the input dataset, helping to improve the effectiveness of monitoring marine oil spill pollution.