EAGE Workshop on Data Science - From Fundamentals to Opportunities

This paper seeks to provide a template for best practices in the successful application of machine learning (ML) methodologies in the E&P business. The objective is to ensure that the benefits of these methodologies are maximized and the pitfalls are avoided. It starts by providing a clarification on the differences and relationships between the two common terminologies: ML and artificial intelligence. Focusing on the machine learning (ML) workflow, which deals with extracting hidden patterns from data and using the patterns to predict various target variables, this paper identifies various opportunities to apply ML and to maximize its value. Some of the challenges facing the implementation of ML in the petroleum industry are discussed. The paper concludes with a few recommendations, professional tips, and best practices to overcome the challenges and avoid the common pitfalls. The outcome of this paper will help to bridge the gap that currently exists between data scientists and domain experts, and provides a collaborative synergy for successful ML projects implementation and maximizing the gains.

Natural Language Processing (NLP) is growing rapidly in this era, fuelled by the advancements in deep learning, the availability of computing resources, and large volume of data. The utilization of NLP for oil and gas has gained significant attention due to its potential in extracting valuable insights from unstructured textual data. However, applying NLP techniques to the oil and gas industry presents unique challenges and pain points that need to be addressed to fully leverage its potential. In this survey, we explore the history, applications, challenges, and future directions of NLP in the oil and gas sector.

The supply chain of the oil and gas industry is known for its complexity and uncertainty. Lasker (2022) conducted a comprehensive study to examine the challenges of the industry’s supply chain and the potential of data analytics to improve its performance. Our own supply chain faces similar difficulties, particularly in predicting Unit of Activity (UoA) such as transaction volume, which currently relies on inconsistent and manual methods due to decentralized data and diverse measurement approaches.

Existing approaches do not provide a systematic way of predicting UoA, and prone to inaccuracies and inefficiencies. To address these issues, this study proposed predictive analytics with a combination of statistical methods and machine learning models to forecast purchase order transaction volume.

According to Lee et al. (2022) , predictive analytics utilizes data, statistical techniques, and machine learning algorithms to predict future events and trends, ultimately improving organizational performance. The proposed methodology follows a defined process, including project definition, data collection, data analysis, statistical analysis, modelling, deployment, and model monitoring.

By embracing predictive analytics, proposed approach offers a more efficient and reliable method for UoA prediction, contributing to optimal supply chain performance.

The increasing demand for legal contract services in today’s business environment requires efficient resource allocation. By leveraging data-driven approaches, organizations can better allocate resources, enhance efficiency, and improve overall performance. This project aims to develop a robust predictive model in forecasting work demand and provide actionable recommendations for resource allocation. Contract volume serves as the demand signal for workload forecasting. Traditional machine learning and time-series forecasting algorithms were assessed, including Random Forest, XGBoost, ARIMA, and Holt-Winters models. Statistical analysis were conducted to estimate the workload volume based on the predicted contract volume. Due to high spectrum of productivity rate, ABC classification method was employed to gain a realistic productivity rate. Integration of data science in the legal function has great potential in unlocking additional insights and opportunities for improvements.

Artificial intelligence (AI) has revolutionized the field of computer science by enabling computer systems to perform tasks that traditionally required human intelligence. As AI rapidly advances, organizations are faced with a critical choice: embrace this technology or risk being left behind in the competitive landscape. Therefore, the democratization of AI has become crucial for organizations, just as important as their day-to-day operations.

AI democratization aims to make AI accessible, inclusive, and available to a wide range of individuals and organizations. Its goal is to create an AI landscape that is inclusive, where people from diverse backgrounds can leverage AI to drive economic growth, address societal challenges, and improve quality of life. By democratizing AI, we strive to promote equity, diversity, and empowerment, ensuring that everyone has access to the transformative power of AI.

This paper presents a strategic framework for AI democratization, emphasizing the systematic efforts required to achieve fair and inclusive access to AI technologies. Key elements of the framework include setting clear objectives, assessing the current data and AI technology landscape, promoting education and skill development, fostering collaboration, establishing ethical guidelines and governance, addressing policy and regulatory considerations, and securing funding and investment.

The Global Business Service (GBS) Digital Data Science team is using Latent Semantic Analysis (LSA) to analyse the massive amount of unstructured text data generated by SLB’s Global Service Desk (GSD). The team is using Dataiku as the project’s data pipeline platform, and the results of various text analytics methods are saved in Azure Synapse for later visualisation with PowerBI. The solution has the potential to improve service desk management, decrease non-value-added activity, identify cost-cutting opportunities, and avoid downtime. The team can identify common issues and develop solutions to prevent them from occuring in the future by analysing the data. The team can use LSA to identify patterns and themes in the data, which can then be used to improve the service desk experience for users. In general, the solution has the potential to improve user experiences, operational efficiency, and decision-making.

With the development of technology in recent years, the amount of data that geoscientists have to work with is increasing in size and quantity. Conventional analytics methods are gradually supplemented by big-data analytics methods and tools, such as artificial intelligence applications, to improve work efficiency and productivity. To collect a large enough set of labels with high confidence to train and predict geological features (such as lithofacies) is time and economically consuming.

In this study, we propose an active learning workflow as a solution for the challenge mentioned above by showing an example of how we improve the quality, quantity and diversity of the lithofacies interpretation. An extensive dataset was used in this study, including: i) 700 000s cuttings samples, ii) 77 000 meters of core, and iii) 70 000s kilometers of logs from 1744 exploration wells in the Norwegian Continental Shelf.

The integrity of onshore pipeline systems is a crucial element for oil and gas companies, including PETRONAS. Geohazards, ground movements, and severe metrological conditions of the environment can cause lots of risk to the onshore pipelines. Natural events including landslides, flooding, slope creep, subsidence, and seismic activities increase the potential for pipeline failure rates throughout the development and operation stages. To address these challenges, a data-driven solution called Integrated Meteorological-Geotechnical System Advisory (iMGESA) has been developed. iMGESA is an in-house risk assessment tool that uses machine learning models to predict the probability of geohazard impact on pipeline failure. It incorporates data from SSGP (Sabah Sarawak Gas Pipeline) Pipeline conditions, rainfall data, and geographical properties to generate outputs including probability of landslide, risk index for geohazard impact to pipeline (target PR), and geopig strain.

The objective of iMGESA is to provide accurate and early intervention on the effect of rainfall on terrain degradation and to prevent further deterioration of geohazard conditions at sites. Additionally, a dashboard system and an online platform have been developed to leverage automation and advanced analytics to generate timely reports on terrain health status at PETRONAS onshore pipeline assets, which staff can access at anytime from anywhere.

The exponential growth in population has resulted in a substantial increase in water use in domestic, agricultural, and industrial sectors. For Malacca state specifically, Syarikat Air Melaka Berhad (SAMB) becomes the main source of water supply for the state population within all sectors. The increase in demand plus a low water supply will results in water shortage issues. Some of the crucial issues faced by SAMB are delayed response and poor control of effluent lead to wastewater plant upsets and non-compliance of final discharge. The unreliable (fresh) water supply has led to disruptions in plant operations disruption which proven to be extremely costly. In addition, utility water systems experience various issues such as contamination, under-performing assets, and frequent leaks.

Water management and sustainability are vital elements in providing a clean and safe water supply for the public [ 1 ]. Thus, to practice these manifestos for Malacca, SAMB teamed up with PETRONAS, to develop a digital solution to predict the water shortage and provide automated prescriptive analytics which will help operators oversee water treatment issues solutions. This solution will help to predict when the next water shortage will occur by employing the dataset collected by SAMB.

This paper focuses on leveraging machine learning approach to optimize fuel consumption in compressor stations in Malaysia. The main objective is to predict the fuel consumption where the model recommends the optimal setpoint, thereby enhancing fuel efficiency and minimizing environmental effect in line with the goal of achieving net zero carbon emissions by 2050. A random forest regression model is used to train the input features that significantly affects the accuracy of the model. Extensive measures and factors were considered during model training in achieving the goal to ease or help operators in the decision-making processes. This study is divided into two processes which is developing regression model to predict fuel consumption and then proceed with optimizing the compressor stations with some identified constraint. The model produced makes a valuable contribution to the gas business sector, exhibiting a high accuracy rate with a correlation coefficient of approximately 95%. Furthermore, the findings demonstrate the potential for substantial cost savings by adopting the machine learning optimization model, with a notable 0.65% reduction in CO2 emissions attributed to fuel consumption over a six-month period.