Machine Learning for Energy Trading, Automotive Sector, and Logistics, presented by BigML's Partners A1 Digital.
Main Conference: Introduction to Machine Learning.
DutchMLSchool: 1st edition of the Machine Learning Summer School in The Netherlands.
Aspirational Block Program Block Syaldey District - Almora
DutchMLSchool. ML for Energy Trading and Automotive Sector
1. Machine Learning for Energy Trading,
Automotive Sector and Logistics
BigML Summerschool, Breukelen, NL
Dr. Dieter Mayr, 8th of July, 2019
2. 2
i. A1 Digital Machine Learning Platform powered by BigML
ii. Our perception on Machine Learning in various industries
iii. Use cases and solutions
iv. Our approach and recommendations
Agenda
3. 3
Who we are
A1 Digital – IoT, ML, Cloud and Security Focus
3 Headquarters
Vienna, Munich and
Lausanne - present
in 10 countries
180
Employees
More than 500
international
customer projects
AffiliatedGroup
A1 Telekom Austria
Group & America Móvil
Your Partner
for Cloud, ML, IoT
and Security
4. 4
A1 Digital‘s ML Team
Our Mission: A1 Digital enables its customers to perform
Machine Learning and Advanced Analytics at Scale
Core ML Team: Data Scientists as
… ML Consultants
… Product Manager
… Data Engineer & DevOps
Our technology stack: one large BigML deployment running
in our own cloud (Exoscale)
Our offer: Ml
7. 7
Market starts demanding ML
How usually discover B2B customers dealing with ML:
SMEs
Little strategy on ML, mainly individual persons
Little low-hanging fruits: they know how to solve their problems
Depending on industry: IT mostly drives efforts for ML
Input often from the higher management „to try ML“
Many obstacles: little resources (time), fear of learning curve,
underestimation of potential of own data, bad first experiences
Large
Enterprises
Centralized vs decentralized ML Teams
Often: strategy + „center of excellence“
Large investments in complex infrastructure
Communication: Business vs. Data Science team
Lack of agile and simple ways to explore ML
10. 10
Data related project remain complex….
• Even with best tools, it remains a
challenge to find a use case with
adequate data…
• ML is just one (important!)
element but cannot solve all
problems
• But: with ML project,
stakeholders learn, understand
the imporance of data and
become creative in finding new
use cases in thair fields.
11. 11
But where we believe BigML is essential creating added value
• Nothing to install or configure
• No programming skills
required
• Smart data input
• Automatic Modeling
• Smart model consumption
PROGRAMMABLE
• Offer basic constructors that enable
sophisticated ML strategies
• API-first
SCALABLE
• Fully-automated infrastructure
• Instant access, instant scale
• All the complexities related to
infrastructure are abstracted away
• Serverless
13. 13
What kind of ML Platform is needed ?
• Nothing to install or configure
• No programming skills
required
• Smart data input
• Automatic Modeling
• Smart model consumption
PROGRAMMABLE
• Offer basic constructors that enable
sophisticated ML strategies
• API-first
SCALABLE
• Fully-automated infrastructure
• Instant access, instant scale
• All the complexities related to
infrastructure are abstracted away
• Serverless
14. 14
Machine Learning for
Energy Trading
REQUIREMENTS
Automated workflow for predicting control energy
prices
Evaluation of historic bids and revenues
OUR SOLUTION
Analysis of historic trading strategy
Expert workshops for defining data sources (spot
market prices, weather, etc.) and feature
engineering
Identify best performing machine learning
algorithms
Dashboard as decision support tool for trading and
evaluating historic bids
RESULTS
10 % higher revenues from auctions
More transparent decisions and reduced workload
15. 15
Energy Trading Dashboard
ML Application for Energy Trading
Download data &
Feature Engineering
Auction
announcements
Filter
announcements
„ML-ready Data“
Price prediction for
the next auction
Auction results
Additional data
16. 16
What kind of ML Platform is needed ?
• Nothing to install or configure
• No programming skills
required
• Smart data input
• Automatic Modeling
• Smart model consumption
PROGRAMMABLE
• Offer basic constructors that enable
sophisticated ML strategies
• API-first
SCALABLE
• Fully-automated infrastructure
• Instant access, instant scale
• All the complexities related to
infrastructure are abstracted away
• Serverless
17. 17
Machine Learning for
Wagon Hire and Rail Logistics
REQUIREMENTS
Forecast model on maintenance time required for each
wagon
Creation of complex data model
Scalable Solution to rapidly analyze vast amounts of data
Open, exportable models, ready to use in production
OUR SOLUTION
Consulting on use case selection
Workshop on data selection (gain vs. efforts)
Review of BigML platform how it fits into their
requirements
Test OptiML on existing challenges
RESULTS
OptiML outperforms existing models
Data Science unit works faster, increases collaboration
Leveraging existing models & efforts (from Python)
18. 18
Bindings
How to use bindings for the ML platform? Python example
Add source file
Create a dataset from source
Split data set into training and
test datasets
Create a model (decision tree)
with a training dataset
Evaluate with a test dataset
Get desired evaluation
parameters
19. 19
Bindings
Use the dashboard to evaluate any steps
Look up the Script-ID
Create execution for new
source with Script-ID
“Scriptify” one step or the
whole workflow
20. 20
What kind of ML Platform is needed ?
• Nothing to install or configure
• No programming skills
required
• Smart data input
• Automatic Modeling
• Smart model consumption
PROGRAMMABLE
• Offer basic constructors that enable
sophisticated ML strategies
• API-first
SCALABLE
• Fully-automated infrastructure
• Instant access, instant scale
• All the complexities related to
infrastructure are abstracted away
• Serverless
21. 21
Machine Learning for
Automotive Supplier
REQUIREMENTS
Central ML Platform enabling potentially thousands
of engineers worldwide to get started with ML
Easy entry in ML with ability to fully scale fast
OUR SOLUTION
Discuss ML efforts and strategy
Develop a PoC (on injection molding machines)
Evaluate results and feedback from subject matter
experts
Consulting to create a concept about how to roll-
out ML to business units
RESULTS
10+ potential use cases
Roadmap, learning program and show cases
Cost-effective strategy for a global accessable ML
platform enabling engineers to optimize data
related routines
22. 22
Challenging market expectations demands development of
ML-led Sense-Predict-React capabilities
• Conformity to
specification
• Product performance
Quality
• Low Rework cost
• High percentage of
passed quality inspection
• Low cost of quality
control
• Delivery Lead Times
• On Time Delivery
• Stock availability
Delivery
• Short production and
delivery lead time
• High accuracy of
inventory status
• High dependability of
internal lead times
• Product Selling
• Competitive Pricing
• Disruption driver
Cost
• Low unit cost of
manufacturing
• Fast inventory turnover
• High capacity utilization
• Product range
• Product portfolio offered
• Volume / product mix changes
Flexibility
• Shortest MRP and set
up times
• Shortest length of fixed
production schedule
• Optimal amount of
operating capacity
24. DataExploration
• Use-Case workshop
• First, quick results
• Follow-up potential
Pilot Package
• Multiple workshops
• Multiple data
sources
• Reliable results
ProductiveSystem
• Improve pilot solution
• Integration in existing system.
• Deployment of Application
Expandon yourown
• Implement further
use cases on your
own
The way to Machine Learning based applications
We enable our customers step-by-step
Continuous Machine Learning trainings and support by A1 Digital experts
from data to results in about 6 weeks
25. 25
Prioritize impact & reuse
Develop ML strategy core analytical capabilities & easy access (MLaaS) platforms
Leverage partner stay focused on your business
Educate (repeatedly) Unleash the Citizen Data Scientist.
Domain knowledge respect expertise and bring ML closer to decision making
Freedom Allow freedom for creativity and potentially failing
Small steps focus on fast first projects and stay agile
Start now
Recommendations to our customers