En platform-drevet fremtid med vær som brensel
Torleif M. Lunde - Chief Product Officer @ StormGeo
Den digitale revolusjonen har ført alle industrier nærmere hverandre, noe som åpner opp for helt nye muligheter. Fra å være et selskap med fokus på best mulig værvarsler, har StormGeo utviklet seg til å jobbe med AI og platformbygging. Sammen med verdens største forsikringsselskaper, teleselskaper, sosiale medier, og shipping-konsern, bruker vi AI som en naturlig del for å effiktivisere deres og egne produksjonslinjer. I foredraget fortelles historien om hvordan StormGeo har tatt været ut i verden, til nå å bygge en platform bestående av hyperlokalt vær, kundedata, AI-dreven beslutningsstøtte, med leveranse gjennom visualisering, APIer, event-drevne prosesser og personlig service.
Torleif Markussen Lunde. Chief Product Officer I StormGeo fra oktober 2018. Ansatt i 2014. Holder en PhD i malaria og klima, og har lang erfaring i big data og data science med erfaring fra olje og gass, transport, forsikring og helse. De siste årene har fokus dreid i retning organisjonsutvikling, strategi og M&A.
Les mer her: https://www.eventbrite.com/e/robotics-og-ai-tickets-55883981493#
6. Freedom to Perform
ABOUT STORMGEO
Owners
59% EQT
26% DNV GL
15% Other
Company
6 global operations centers
with 24/7/365 support
2 data centers
27 offices
15 countries
410 employees
Business
66% Shipping
13% Oil & Gas
10% Renewables
8% Cross Industry
3% Media
SHIPPING
12,000
vessels supported in total
29%
penetration (SOLAS) with a
StormGeo service
OIL & GAS
90 000+
Unique customer specific
forecasts issued every month
43%
Market share in North Europe
18%
market share world wide
RENEWABLES
30%
Market share in offshore wind
worldwide
7%
Market share in electric utility
operations in the U.S.
CROSS INDUSTRY
11 000+
U.S Onshore locations served on
a daily basis
Business Continuity
solutions used by a variety of
industries to enhance safety
and minimize downtime
MEDIA
We boost your ratings
Global weather data and
professional visualization across
any device or screen.
Geography
main portion of customers
served in Europe, Middle East
and Asia.
7. • Principal component regression, Software
Linpack
• Quantile regression, Software R
• Logistic regression, Software R
• Bayesian Statistics, Software R
• +++
2004-2011
ADVANCED STATISTICS
• Neural networks, software pybrain
• Support Vector Machine, software Scikit-Learn
• Optical Flow, software OpenCV
• Agglomerative Clustering, software Scikit-Learn
• Random Forest Classification, software Scikit-Learn
• Extreme gradient boosting, software xgboost
2011 - Present
MACHINE LEARNING
• Multiple linear regression, software S
2003
MODEL OUTPUT STATISTICS
• Various neural network architectures, software
Keras with Tensorflow backend
• Running deep learning on four NVIDIA Titan X
GPUs
2016-present
DEEP LEARNING
Siden 2003 har vi vært på søken etter den perfekte morgendagen
Data
Science
8. Hele forretningen til StormGeo er bygget rundt data, prediksjon og kommunikasjon
Freedom to Perform
10. Weather is the major
cause of disruptions to
society globally
2017: US $340 bn in overall
losses due to natural disasters
(*) Source: Munich Re
Freedom to Perform
(*)
11. Mobilitetsdata + smart meters + hyperlokalt vær + AI = strømforbruk
Human mobility
Weather
Electricity
Prescriptive model
Forecasted house
level electricity
consumption
12. Weather
Holidays
Macro economy
Light hours
Market events/UMMs
Input
∞Price
Production
Forecasting power consumption is key to plan power production, taking into
account production from different sources in different areas. To improve our
forecasting of power consumption, we combined weather intelligence with local,
real time and historical power consumption data, to build a neural network going
into our power price forecasts for the Nordic Power market. As new data becomes
available, the model continue to improve.
En modell for nordiske og europeiske stømpriser
Observed
Day ahead forecast
(2% error)
Real time power consumption forecast
13. 61 Categorical exposure features
• Building material
• Roof type
• Line of Business
• Company
• Client age group
8 Numerical exposure features
• Age of home/roof
• Size of home
• Client previous losses
• Client insurance score
• Client non-payments
2 Weather radar features
• Max radar reflectivity
• Average radar reflectivity
50 Wind features
• Wind and gust
• 5 different severity percentiles
• 5 different geographical areas
121 Input features for
model
1 Output feature
Likelihood of claim
Weather + Exposure = Claims
Machine learning architecture
14. PAST
What just happened?
PRESENT
What is about to happen?
FUTURE
What will happen soon?
En av verdens største forsikringsselskaper
bruker StormGeo for å varsle hvilke hus har høy
risiko for skade
15. The StormGeo Infinity differentiator
Normalization of data is the key to valuable performance analysis
Comparing vessel performance
and analyzing vessel trends using
speed over ground can be highly
misleading
Speed over ground differences is
first of all connected to
differences in weather at the
given time
To analyze performance
differences, the effect of the
weather and currents on vessel
speed needs to be subtracted
The same goes for loading
conditions and several other
parameters
The data needs to be normalized
such that one is left with only
analyzing based on the relevant
performance speeds
Speed Over Ground
Raw Data
Laden
Ballast
Consumption
Speed Over Ground
Performance Speed
Normalized Data
Laden
Ballast
Consumption
Performance Analysis
Sister vessel trade analysis
Sister vessel Fuel consumption
analysis
Fleet Transport Fuel EfficiencyPerformance Speed / Calm Sea Speed
16. En platform-drevet fremtid med vær som brensel
Vortex SaaS
• Business optimization for
weather exposed industries
• Digital customer experience and
intimacy on desktop and mobile
• Delivery and interaction SaaS for
data science driven Weather
Insights
BUSINESS
ECOSYSTEM
</Programmable APIs>
• Data platform allowing
integration in external
enterprise-scale ecosystems
• 99.98% uptime in 2018
• The connective tissue in
StormGeo’s digital ecosystem
Event driven
• High-definition, IoT, micro-
weather information as key
enabler of autonomous data
driven operations
• Reactive, event driven, mobile
notifications
Data Science & Personal Service
• 24/7 Weather Insights
• Tailored solutions by
customization and integration of
third party data
• Safe and SOX-compliant data
platform
• Personal users and role
management allowing super
users and self service
17. Working in close partnership with StormGeo;
Data science in the core of your value creation
Turbulence
Workshop – when
you know your
workflow can be
improved, but you
don’t know how
Pilot – when you
know what to
improve, but need
expert help
executing
• Feasibility
assessment
• Roadmap
development
• Organization
education
• Data validation
• Feasibility
assessment
• Business
development
and
identification of
stakeholders
Discovery: 2-4 weeks
Data Science
Business Development
Breeze
Data collection & refining
Company data is processed, cleaned,
refined and restructured
Method selection and exploration
Based on the business objective and
data, a customized solution to solve the
specific problem is designed
Prototype deployment
Proof of concept, validation of
methodology and estimated accuracy of
the selected approach
Data Science
Proof of Concept: 1-4 months
Gale
Product testing and implementation
Test with users to confirm benefits
Site agnostic implementation
Implementation of solution on or off
premise
Value creation through delivery method
Deliver operational system with
visualization and APIs.
Data Science
UX, IT and Front-End Developers
Business Development
Production: Project specific
Storm
Operational stability
Support or take full responsibility of the
operational product
Validation and maintenance
Follow up with validation and calibration
if data input changes
Data Science
IT, Support
Reinforcement and stability
DeepStorm