Can agent-based modelling help to update conceptual geological models? - A fractured reservoir example.

Bastian Steffens and Vasily Demyanov and Dan Arnold and Helen Lewis. ( 2021 )
in: 2021 RING Meeting, ASGA

Abstract

In this work we apply agent-based modelling (ABM) to assist with updating prior conceptual geological models in uncertainty quantification studies. Conceptual geological models allow geoscientists to integrate available subsurface data with their geological understanding and make predictions about rock property distributions in the subsurface . The geological concepts are then used to guide the design of reservoir models for fluid flow simulations. Reservoir models and fluid flow simulations are required to assess of the feasibility of potential hydrocarbon or geothermal energy production sites. Commonly such reservoir models rely heavily on geological conceptual models to partially compensate for severe hard data shortages as collecting new data is expensive and time consuming. Any such reservoir model is subject to updates as new information arrives throughout the operational lifetime of a reservoir and tends to cause reassessing and overworking of the conceptual geological model. We combine agent-based modelling with flow diagnostics, a physics-based proxy for reservoir dynamics, to iteratively update conceptual geological models to fit newly incoming production data. This enables fast screening of a range of conceptual model scenarios. In agent-based modelling, autonomous agents simulate complex adaptive systems by following a set of simple predefined rules when interacting with each other and their environment. Each agent's behaviour can be goal-directed, allowing that agent to compare the outcome of its behaviours against its goals and adjust its responses accordingly. Flow diagnostics represent simplified numerical flow experiments that can be understood as proxies for full physics fluid flow simulations and run within seconds. Output metrics, such as the time of flight, indicate the swept and unswept areas of the reservoir model. In our case, the agent's environment is a 2 dimensional grid that represents the initial conceptual geological structural model that determines the character of the fracture pattern distribution. The agents compete against each other for areal occupation of this environment. Agents are represented as points in the environment and subdivide the environment among themselves via Voronoi tessellation. For the area it occupies, each agent selects from a range of facies and petrophysical property distributions taken from the specific location in the underlying geological conceptual model. Agents move around in the environment and adjust their porous properties to balance two major rules: (i) minimising the misfit between the reservoir models flow diagnostic response and the production data and (ii) selecting petrophysical properties that honour the underlying geological conceptual model. If no satisfactory balance is found, agents can update the pre-existing underlying geological conceptual model in their favour, allowing them to pick from different property distributions. This information can then be used to reassess and update the conceptual geological model and build new reservoir models accordingly. This agent-based modelling technique is tested on a synthetic fractured reservoir designed with the Teton anticline's (Western USA) conceptual model in mind.

Download / Links

BibTeX Reference

@inproceedings{STEFFENS_RM2021,
 abstract = { In this work we apply agent-based modelling (ABM) to assist with updating prior conceptual geological models in uncertainty quantification studies. Conceptual geological models allow geoscientists to integrate available subsurface data with their geological understanding and make predictions about rock property distributions in the subsurface . The geological concepts are then used to guide the design of reservoir models for fluid flow simulations. Reservoir models and fluid flow simulations are required to assess of the feasibility of potential hydrocarbon or geothermal energy production sites. Commonly such reservoir models rely heavily on geological conceptual models to partially compensate for severe hard data shortages as collecting new data is expensive and time consuming. Any such reservoir model is subject to updates as new information arrives throughout the operational lifetime of a reservoir and tends to cause reassessing and overworking of the conceptual geological model. We combine agent-based modelling with flow diagnostics, a physics-based proxy for reservoir dynamics, to iteratively update conceptual geological models to fit newly incoming production data. This enables fast screening of a range of conceptual model scenarios. In agent-based modelling, autonomous agents simulate complex adaptive systems by following a set of simple predefined rules when interacting with each other and their environment. Each agent's behaviour can be goal-directed, allowing that agent to compare the outcome of its behaviours against its goals and adjust its responses accordingly. Flow diagnostics represent simplified numerical flow experiments that can be understood as proxies for full physics fluid flow simulations and run within seconds. Output metrics, such as the time of flight, indicate the swept and unswept areas of the reservoir model. In our case, the agent's environment is a 2 dimensional grid that represents the initial conceptual geological structural model that determines the character of the fracture pattern distribution. The agents compete against each other for areal occupation of this environment. Agents are represented as points in the environment and subdivide the environment among themselves via Voronoi tessellation. For the area it occupies, each agent selects from a range of facies and petrophysical property distributions taken from the specific location in the underlying geological conceptual model. Agents move around in the environment and adjust their porous properties to balance two major rules: (i) minimising the misfit between the reservoir models flow diagnostic response and the production data and (ii) selecting petrophysical properties that honour the underlying geological conceptual model. If no satisfactory balance is found, agents can update the pre-existing underlying geological conceptual model in their favour, allowing them to pick from different property distributions. This information can then be used to reassess and update the conceptual geological model and build new reservoir models accordingly. This agent-based modelling technique is tested on a synthetic fractured reservoir designed with the Teton anticline's (Western USA) conceptual model in mind. },
 author = { Steffens, Bastian AND Demyanov, Vasily AND Arnold, Dan AND Lewis, Helen },
 booktitle = { 2021 RING Meeting },
 publisher = { ASGA },
 title = { Can agent-based modelling help to update conceptual geological models? - A fractured reservoir example. },
 year = { 2021 }
}