Autonomous agents

Multi-agent systems à la ‘The Sims’ trialed by ExxonMobil in production optimization. Pump, well and facility agents cooperate and conflict to mimic real world scenarios.

Speaking at the SPE/Reed Exhibitions Intelligent Energy event earlier this year, ExxonMobil’s Mike Romer reported on trials of ‘autonomous agents’ (AA) in oil and gas asset management. Autonomous agents, combined into a ‘multi agent system’ (MAS) are software tools for modeling complex, multi-variable systems where an analytical approach is impossible. A popular example of such a tool is ‘The Sims’ computer game which uses the approach to model a virtual world. Romer cited AA guru Michael Wooldridge who defines an agent as being ‘capable of autonomous action in its environment with delegated goals.’ Wooldrige sees agents as a ‘next generation’ programming paradigm ‘beyond objects.’

An asset such as an electric submerged pump can be represented as a pump management agent with certain ‘needs’ such as over current shut off and other constraints. Agents may have conflicting goals. A well management agent may attempt to speed up a pump while the pump agent is trying to avoid emulsion creation. A surface facility agent may not have the capacity needed for the extra fluids.

Exxon’s research includes workflow automation using the Prometheus methodology to inform agents with knowledge from subject matter experts with a wide range of skills. The approach involves scenario development, agent role definition and the development of an analytical model and knowledge-base.

The degree to which a workflow is automated can vary. Some processes will be under automated (closed loop) control. Others, particularly where objectives conflict, may involve human intervention—leveraging model-derived ‘low level’ tasks that can be performed unambiguously. The software agents are intrinsically modular and can be combined and re-configured for use in different circumstances.

Romer observed that ‘workflow modeling at the granularity needed for automation is challenging for real-world applications.’ The full paper is available from the SPE. For more on intelligent agents read the introduction by Ira Rudowsky. AA/MAS appears to be gaining traction in academia with a well attended conference in Valencia, Spain this month. Romer also cited early application of the technology for Statoil by researchers at Norway’s NTNU. Commercial software (Jack) is available.

Click here to comment on this article

Click here to view this article in context on a desktop

© Oil IT Journal - all rights reserved.