Professional Documents
Culture Documents
Artificial Intelligence
Comes of Age in Oil and Gas
Gentry Braswell, JPT Online Technology Editor
51
Online
Intelligent Automation
Actual setup pyramid
Assistant
System
Other
stakeholders
Onshore
ERP ERP
planner Strategic control layer
Strategic decision
layer
Offshore Tactical control layer
Tactical decision production
layer engineer MES
Process systems
A model of the software architecture in use at the Siri area offshore Denmark. (Image courtesy of Dansk Olie og
Naturgas E&P.)
data through AI, for example enabling nonproductive time during drilling oper- behind the natural language. In our tools
the optimization of fracture spacing and ation, as well as real-time data analysis it is applied in the pattern recognition
fracture design models. as applied to full-field modeling in smart function of the software,” Mohaghegh
Within a period of 6 months, field applications. said. “Artificial neural networking is the
Mohaghegh’s AI-based reservoir mod- The reservoir management soft- key to learning those patterns that have
eling software performed a 135-well his- ware, for example, functions in three been discerned.”
tory-matching analysis of the Marcel- stages: data mining, data-driven model- Evolutionary algorithms determine
lus shale, “which is unheard of because ing, and model analysis. First, the data, patterns that are relevant for the par-
numerical models that are built for shale which is often hyperdimensional, is pre- ticular field being modeled, by making
are overwhelmingly single well models processed for quality control, and is predictions, evaluating those predic-
and are computationally very expen- mined to discover hidden and potential- tions, then discarding incorrect predic-
sive,” Mohaghegh said. The power of ly useful patterns. tions. “Genetic algorithms are rooted in
pattern recognition through AI is what The first step continues with iden- Darwinian evolution theory. The same
enables this kind of speed in history tification of key performance indicators natural approach to human evolution
matching. The modeling software is (KPI) where the software determines is applied to fracturing a well,” Moha-
a hybrid system, combining artificial the parameters that controls the out- ghegh said. “It is a learning algorithm,
neural networks, fuzzy set theory, and come. The relevant KPIs vary from field an ensemble.”
genetic algorithms. to field, and there can be hundreds and In the second step, the data is
ISI focuses on reservoir and pro- hundreds of indicators for consideration parsed into machine-usable form, which
duction management, optimization of on a given project. renders the reservoir model. The model
hydraulic fracturing, reservoir simula- Fuzzy logic serves the pattern rec- is developed based on measured data
tion and modeling, and analysis, pre- ognition function of the software. “Fuzzy from the field with little to no interpre-
diction, management, and reduction of logic is multivalued logic. It is the logic tation involved. “We let the wells and
the reservoir speak for themselves and optimization. Unlike the more academ- fering with safety level. We are just add-
impose their will on the model, instead ic and speculative nature of reservoir ing a level on top of existing control
of imposing our current understanding modeling with AI, oil and gas produc- systems,” Mikkelsen said. Today’s con-
of the geology and physics on the model. tion optimization software gets closer to trol systems should be used as the input
The model is then validated by testing delineating the real-time risks of critical source for this system, and configura-
it with blind data during post-modeling production operations. tion is an issue that should not be a
analysis,” Mohaghegh said. Operators of complex installa- problem, he said. “Software interfaces
“Data always comes from a specific tions must focus on many tasks. Forget- and programs exist, though older soft-
field, a specific model, a specific situa- ting to start a pump, for example, may ware tools could be more troublesome,”
tion. It is always field by field,” Moha- result in less efficient production. Mul- he said.
ghegh said. “Our software is a workflow tiple-objective control software in use The stratified multi-objective con-
that takes you from the data you have at a pilot project in the Siri area offshore trol software is extensible, sizable, and
for a specific operation to a data-driven Denmark allows a level of operational can be run on a single processor or in
Top drive
Drillstring
A block diagram of the main components of Robotic Drilling Systems’ robotic drill floor.
(a) (b)
A 3D field model, left, and a 2D semisynthetic model of the Norne field, using particle swarm optimization for
estimating porosity and permeability from time lapse seismic and production data. (Image courtesy of Norwegian
University of Science and Technology’s Integrated Operations Center.)
decision making), through multiagent the system,” Mikkelsen said. “We can let genetic algorithm deployed in the soft-
software technology, and then using them run in parallel in small control con- ware makes it customizable, so the oper-
the genetic algorithm for the commu- texts, which is very important, so we can ator can add and remove new concerns
nication between the elements,” Mik- add, update, and remove them from the on the fly, and an offshore planner can
kelsen said. The project is described running system.” just use the parts of the system relevant
in technical papers SPE 156946 and A multiagent system approach to their project. There can be any num-
SPE 153815. seems also to solve the challenge of rigid ber of parameters incorporated, so the
“We can right away increase pro- architecture and inflexibility of produc- system is applicable on very large, com-
duction throughput. This is a step toward tion software, he said. “Most modern plex fields.
integrated operations because you have control software is fixed and quite stat- “The oil and gas domain has a ten-
to make all your objectives explicit. We ic. When we call a vendor, the slightest dency to become more and more com-
have an agent for each concern, and they change in a control system today is very plex. Operators cope with this challenge
are active all the time, for each part of expensive,” Mikkelsen said. However, the of complexity all the time,” M ikkelsen
said. “This is a way of approaching all to operate fully unmanned,” said Roald demonstrated several hundred times
these new fields with deeper water and Valen, control system manager at RDS. since then.
greater volumes of data. And matur- “We knew from the traditional industry “If you go from manual work to auto-
ing fields as well. We have also a lot where robots are being used in automat- mated work, then the next step is going
of brownfield applications, and even ed systems that without the autonomous for complex automated systems. Today
the brownfields are getting more and abilities, there would be a lot of tweak- our drill floor mainly consists of remote-
more complex.” ing and tuning of the system to get it ly operated machines requiring humans
right. Together with the fact that on the to supervise and control the activities
Drill Floor Automation drill floor there are too many variables and therefore also be the key factor in
In September, Norway’s Robotic Drill- changing, we knew that it would be close regard to the performance of the drill
ing Systems (RDS) signed an agree- to impossible to get it automated, so floor,” Valen said. “A good team can per-
ment with the United States Nation- we had to include autonomous abilities. form great while other teams might do
al Aeronautics and Space Administra- We contacted NASA and the Jet Propul- a poor job, but for any processes involv-