Built-in machine learning and big data analysis method help prevent events that could lead to flaring


The United Nations 2030 Agenda for Sustainable Development, presented in New York in September 2015, identifies 17 Sustainable Development Goals (SDGs) which represent common goals for today’s complex challenges and constitute an important reference for the international community.

One of the global priorities is to fight climate change, aiming for zero process flaring by 2025 and reducing emergency flaring caused by hazardous events. An integrated machine learning and big data analytics framework has been developed to prevent and manage hazardous events that can lead to emergency flaring.

This framework’s ability to address and manage hazardous events in advance or in real time provides field engineers and operators with critical support in identifying operating parameters that need to be managed quickly.

This article presents the development and testing of a method to predict upstream events that could lead to flaring, by applying an integrated framework. The central idea is to leverage machine learning and big data analytics to manage major disruptions that would lead to inefficiency and big losses. The tool is developed for complex upstream production systems, where a disruption can be caused by many different factors, leveraging data-driven monitoring systems to identify weak signals of upcoming events.

The proposed framework is mainly composed of a pipeline divided into three modules operating before, during and after an event. The first aims to reduce the probability of an event, the second works on the severity and the third has a double function: to report disturbances and to collect feedback to be used to improve analyses.

The predictive component alerts operators when it recognizes a dangerous pattern among the parameters considered. The other two components can support the first and can be exploited to detect early signs of deviations from the proper operating envelope that the predictive component does not detect. In addition, during an event, operators can quickly identify the causes of the disturbance. This allows a faster reaction and, therefore, a significant reduction in amplitude. The proposed solution offers the following two complementary methodologies:

  • An agnostic anomaly detection system, allowing to map abnormal behaviors as a dynamic operating envelope and identify the most affected units
  • Real-time root cause analysis as a vertical solution, with learning gained from monitoring different units

The tool can also provide automatic event logging using information provided by the root cause system, including operator feedback, which will improve the performance of each module in the framework.

The entire pipeline was applied online, working with real-time data from an operating oil field, with a particular focus on the purge and flare systems. The generated architecture is able to overcome some of the main problems related to the complexity of upstream production assets, such as lack of data, rapid changes in physical phenomena and randomness of disturbances. The first test demonstrates that the tool’s accuracy identifies and suggests actions on 35% of the most dangerous flaring events.

Download the full article from SPE’s Health, Safety, Environment and Sustainable Development Technical Discipline page free of charge until April 20.

Find SPE paper 200942 on OnePetro here.


About Author

Comments are closed.