Integrated Machine Learning, Big Data Analytics Method Helps Prevent Events That Could Lead to Flaring
This paper presents the development and test of a method to predict upstream events that could lead to flaring, applying an integrated framework using machine-learning and big-data analytics.
The United Nations’ 2030 Agenda for Sustainable Development, presented in New York in September 2015, identifies 17 Sustainable Development Goals (SDGs) that represent common goals for the current complex challenges and are an important reference for the international community.
One of the priorities at a global level 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 was developed to prevent and manage the hazardous events that can lead to emergency flaring.
The ability provided by this framework to tackle and manage in advance or real-time hazardous events gives field engineers and operators critical support to identify operating parameters that must be managed rapidly.
This paper presents the development and test of a method to predict upstream events that could lead to flaring, applying an integrated framework. The core idea is to exploit machine-learning and big-data analytics to manage major upsets that would lead to significant inefficiency and loss. The tool is developed for complex upstream production systems, where an upset could be caused by many different factors, exploiting data-driven monitoring systems to identify the weak signals of upcoming events.
The framework proposed is mainly composed of a pipeline divided into three modules operating before, during, and after an event. The former aims to reduce the probability of an event, the second works on the severity, and the third one has a dual function: to report upsets and to gather feedback to be used to improve analytics.
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. Moreover, during an event, operators can promptly identify the causes of the upset. This allows a faster reaction and, consequently, a significant reduction in magnitude. The proposed solution provides the following two complementary methodologies:
- An agnostic anomaly detection system, helping to map anomalous behavior as a dynamic operating envelope and identifying the most affected units
- A real-time root-cause analysis as a vertical solution, with learning obtained from monitoring different units
The tool also can provide an automatic event register using information provided by the root-cause system, including operator feedback, that will improve the performance of each module of the framework.
The entire pipeline has been applied online, working with real-time data coming from an operating oil field, with a special focus on blowdown and flaring systems. The architecture generated is able to overcome some main issues related to the complexity of upstream production assets, such as a lack of data, rapid changes of physical phenomena, and randomness of upsets. The first test demonstrates that the tool accuracy identifies and suggests actions on 35% of the more-dangerous flaring events.