The complexity, criticality, and real-time demands of the energy sector make it a prime candidate to benefit from applying machine learning. This session presents two case studies of machine learning automating decisions for energy companies.
For the largest windfarm operator in North America, machine learning applies predictive and prescriptive analytics to the complex task of scheduling crews for maintenance and repairs. Automating the scheduling process across multiple windfarm sites saves the operator millions in labor costs per year and frees managers and crews to do actual work. Machine learning also evaluates ever-changing conditions and automatically reschedules workers and tasks as necessary.
For a large European energy company, online machine learning provides a systematic and automated approach to commodities trading, including creating and executing trading strategy and predicting prices.
2. What We Do
World-Class
Platform
We develop advanced analytics applications for asset-
intensive industries that focus on operations in context, in
motion, and in real time: collecting and correlating data,
predicting asset condition and operation, optimizing and
automating networks of people and assets, and visualizing
everything in real time.
Our applications are developed on our own advanced,
end-to-end platform, and our machine learning engine
produces intelligence agency-class analytics. This gives
us a huge advantage in quality and time-to-value for our
customers.
Analytics for the IoT Revolution
3. Why SpaceTime?
Rapid time-to-value and
quickly add new
capabilities
Continuous
optimization
and action at the edge
Past and Future
Proof
Connect legacy,
multi-vendor, future
Predict and optimize
under uncertain conditions in
real time
4. We’re Leading the Revolution
SpaceTime has a global, Fortune 500 customer base spanning key industrial
markets, including 8 of the 20 largest utilities.
Key Customers Global Presence
Singapore
Japan
EuropeUK
Canada
Corporate HQ
Silicon Valley
6. - Pricing Optimal Battery Warranty
- Commodities Trading
- Case Study Predictive & Prescriptive
Windpark Maintenance
7. Use Time Series To Predict Nonlinear Battery Failure
Innovative machine learning to find hidden
patterns in time series
Predict capacity degradation without having
seen it in the wild
Combine Hidden Markov Model with
Hierarchical Mixture Models
Why
Reduce accruals by reducing
financial risk of warranties
How
Predict degradation over time
Find optimal policy for
warranties
8. Failure Time Prediction To See Over The Hill
The failure time prediction
gives the probability of failure
into the future –
at any given point in time
9. CBM with
Prediction
Assumes failure at A:
lower asset life, increased
repair/replacement costs
Time in Operation
Advanced Machine Learning For Optimized Asset Lifecycle
Failure
SpaceTime
Machine Learning
● Predict forward in time
without loss of confidence
● “See over the hill” to extend
asset life
● Produce better optimization
for lower costs and
increased productivity
A
Optimization
Zone
Probability of
Failure
Predict Failure
Optimize
Operations
Detect
Anomalies
Extended Asset Life
B
10. Comparing Predictive Models
Risk tolerance threshold for
unit failure, e.g. 12.5%
• Risk tolerance is a constraint
on operations
• The model that produces
fewest false negatives the
longest is the most efficient
• Accuracy is not the most
important metric!
• The trade off between failure
rate and operating hours is
more meaningful
12. The Holy Grail Of Time Series Beating The EMH
Innovative machine learning to find hidden
patterns in time series
Combine deep learning with hierarchical
dynamic Bayesian Models to predict price
Use stochastic optimization for money
management
Learn optimal trading policy
EMH
Efficient Market Hypothesis
Market reflects all relevant
info
Systematics Trading
Make money beating the
EMH
13. Futures Contracts
Energy
BRENT CRUDE
WTI CRUDE
US NATURAL
GAS
UK NATURAL
GAS
ETHANOL
Brent CrudeMetals
GOLD
SILVER
COPPER
PALLADIUM
Crops
CORN
WHEAT
SOYBEANS
OIL
SOYBEANS
MEAL
14. Helping Largest Wind Farm Operator
Make Decisions Under Uncertainty
• Reduced crew hours: $2.3 million
savings/location
• Optimized crew schedule
• Improved crew safety and regulatory compliance
• Solution – Crew Optimization – 250 users
Success Story: Predictive Maintenance & Optimization
● Largest wind farm operator in the world; 19
states and 4 Canadian provinces
● 100+ sites; 10,000+ turbines; 1,000
teammates
“Using advanced analytics to
optimize resources and efficiency
allowing us to reclaim thousands of
lost hours of productivity”
General Manager
Largest Windfarm Operator Energy Resources
15. Optimization
Weather Forecasts
Crew Availability
Work Order List
Sensor
Data
Crew Schedule
Crew Route
Work Order
List
Traffic
Value of Activities
Performed
Risk and Cost
Managed
Crew Skills
Asset Failure
Model
Other Models
DATA INPUTS OUTPUTS/ACTION
Remaining
Useful Life
SpaceTime can perform optimization even when
inputs involve uncertainty, like weather or
traffic, and constantly changing inputs like the
probability of asset failure.
Global Optimization of Operations
16. Reasoning Under Uncertainty Over Graphs
Speech Recognition Computer Vision
Assets As ApplicationsGames
Time in
Operation
Failur
e
A
Optimiza
tion
Zone
Probability
of Failure
Predict
Failure
Optimize
Operation
s
Detect
Anomalies
Extended Asset Life
B