Machine Learning in Oil and Gas
Dave Lafferty
Introductions
Analytics Framework
What Is Machine Learning
Branch of artificial intelligence
Learns from training examples to predict future events
Mimics human reasoning
Best used for when there is a complex relationship between variables
Methods include Neural networks
Multilayer perceptron (MLP)
Radial basis functions
Support vector machines
Naïve Bayes
k-nearest neighbors
Geospatial predictive modeling
Why Machine Learning for Oil and Gas?
Ability to Scale
Ability to See More Complex Relationships
Ability to Automatically Continuously Refine Models
Yielding
Increased Production/Productivity
Reduced Operating Costs
Enhanced Safety
Barriers To Machine Learning
Machine learning can significantly increase production,
reduce costs and improve safety
However
A whole new approach needs to be used that goes
around the cost and complexity of conventional systems
to over come key barriers to adoption
Barrier 1: Getting Data Into Shape
Sensor Networks
Historians
Operator Logs
Control SystemEvents
Encryption
Data Ingestion
Data Cleansing
Normalization
Contextualization
Aggregation
Segmentation
Anonymization
Data Lake
Barrier 2: Moving Beyond Simple Analytics
Descriptive – What is happening now based on incoming
data. To mine the analytics, you typically use a real-time
dashboard and/or email reports.
Diagnostic – A look at past performance to determine what
happened and why. The result of the analysis is often an
analytic dashboard.
Predictive – An analysis of likely scenarios of what might
happen. The deliverables are usually a predictive forecast.
Prescriptive – This type of analysis reveals what actions
should be taken. This is the most valuable kind of analysis
and usually results in rules and recommendations for next
steps.
Descriptive Analytics Issues
Tells you what has happen – not
what will happen
Difficult to scale – people
interpret the information
“What” leads to “Why”
Difficult to pick the right KPIs
Tend to be static reports
Diagnostic Analytics Issues
Tends to be very laborious
Limited amount of “actionable”
insight generated
Provides good insight to a very
specific part of a problem
Cannot handle complex problems –
best for simple cause and effect
Barrier 3: Providing An End to End Solution
Sensing
Communicating
Big Data
Analytics
Visualization
Value
Barrier 4: Lack Of Domain Knowledge
Results must be put into actionable form
90ft case
Results must be merged back into 1st principals
Data driven models must meet the physical world
Barrier 5: Overcoming Machine Learning Issues
Requires clean data – data prep maybe
90% of the effort
Models require continuous refinements
Cloud based analytics better suited for
high latency applications (predictive
maintenance)
May require large amounts of data
streamed to the cloud
Barrier 6: Articulating Business Value
Think about business problems – not as a data science
problem
Present as verticals – not generic solutions
Production optimization
Corrosion
Chemical management
Equipment maintenance
Equipment health
Completions optimization
New Developments
New Business Models
“Behind the Firewall”Conventional Licensing
Infrastructure As a ServiceIaaS
Software As a ServiceSaaS
Platform As a ServicePaaS
Data As a ServiceDaaS
Product As a Service“Power By The Hour”
Owner/Operator
Risk
SharedRisk
SupplierRisk
New Developments
Edge Computing Edge Sensors – storage, computing and
communications
“Fog” computing
Low Power Wide Area (LPWA) Public Networks More like cellular service
Ubiquitous coverage
Shared infrastructure -> lower costs, faster time to market
Watch RPMA. LoRa and Sigfox
Open Automation Promoted by ExxonMobil
Make systems look like more aircraft systems than DCS
Smart RTU
Control
Loop
Data
Optimization
Model
Fog Computing
One or more collaborative end-user clients to carry out a
substantial amount of storage, communication, control,
configuration, measurement and management directly in the
field
Conclusions
Consider Two Things During The Conference
1) How Can I Use Machine Learning To Bring Value To My
Organization?
2) How Can Overcome The Barriers To Adoption?