Deep Learning for Automated Process Discovery: A Case Study of the Oil and Gas Industry
The oil and gas industry (OGI) has always been associated with challenges and complexities. It involves many processes and stakeholders, each generating a huge amount of data. Due to the global and distributed nature of the business, processing and managing this information is a challenging task.
Many issues such as orchestrating different data sources, owners and formats; verifying, validating and securing data streams as they move along the complex business process pipeline; and getting insights from data for improving business efficiency, scheduling maintenance and preventing theft and fraud ought to be addressed. Artificial intelligence (AI), and machine learning (ML) in particular, have gained huge acceptance in many areas recently, including the OGI, to help humans tackle such complex tasks.
Furthermore, multi-agent systems (MAS) as a subfield of distributed AI meet the requirement of distributed systems and have been utilized successfully in a vast variety of disciplines.
Several studies have explored the use of ML and MAS to increase operational efficiency, manage the supply chain and solve various production- and maintenance-related tasks in the OGI. Blockchain, a potential emerging technology has also been suggested for the oil and gas sector, it is expected to provide a reliable way of conveying oil and gas down the supply chain, to ensure the safe delivery of products, and provide a transparent system for all participants.
All aforementioned approaches have high potentials in the oil and gas sector, especially when they are combined. While there were a few studies exploring the application of these technologies to OGI, not many, if any, were implemented in real life. This is due to lack of funding, various real-time constraints, shortage of development tools fit for efficient industrial implementation, risks associated with the acceptance of new technologies not confirmed as working or beneficial in big industries. A more likely approach may be process mining, an approach that is actively being used by many businesses for understanding and improving complex business processes.
Process discovery, a starting point of process mining has been extensively studied, with several proposals for automated process discovery methods. However, there are still some limitations in the state-of-the-art methods. For example, they produce large and spaghetti like models or models with poor fitness and precision (i.e. inability to discover process models that will express observed behaviour in the best possible form). It has proven difficult to attain a compromise between the four process discovery quality metrics (i.e. fitness, precision, generalisation and complexity).
To address these issues, this research introduces a deep learning approach for automated process discovery to improve the accuracy and visual presentation of models. The new method will be evaluated on a collection of real-life oil and gas event logs. The new process discovery method will be compared based on several metrics such as fitness, precision, generalization and complexity to other popular state-of-the-art automated process mining methods.
