Upgrade to Pro
— share decks privately, control downloads, hide ads and more …
Speaker Deck
Speaker Deck
PRO
Sign in
Sign up
for free
Machine Learning and Simulations for Asset Management at Alliander
Gijs Molenaar
March 13, 2020
Science
1
36
Machine Learning and Simulations for Asset Management at Alliander
Gijs Molenaar
March 13, 2020
Tweet
Share
More Decks by Gijs Molenaar
See All by Gijs Molenaar
gijzelaerr
1
35
gijzelaerr
3
86
gijzelaerr
0
39
gijzelaerr
1
48
gijzelaerr
1
68
gijzelaerr
1
92
gijzelaerr
0
120
gijzelaerr
3
190
gijzelaerr
0
56
Other Decks in Science
See All in Science
taichiiiiiiiiiiiii
0
660
usamik26
1
940
kamakiri1225
0
590
konakalab
0
310
shomaekawa
1
140
usamik26
0
800
hachama
0
140
shuntaros
0
250
masayamoriofficial
0
520
kanaugust
PRO
0
100
tomohideshibata
5
2.5k
ymdpharm
0
460
Featured
See All Featured
phodgson
87
3.9k
davidbonilla
70
3.5k
kneath
219
15k
roundedbygravity
84
7.8k
stephaniewalter
260
11k
revolveconf
200
9.6k
searls
204
35k
jakevdp
774
200k
malarkey
393
60k
reverentgeek
168
7.1k
ddemaree
274
31k
tmm1
61
8.4k
Transcript
Machine Learning and Simulations for Asset Management March 13, 2020,
NL-RSE Meetup Gijs Molenaar
pythonic.nl • Me! • Freelance Software/Research Engineer • MSc AI
(UvA) • PhD astronomy (South Africa)
• Public Utility company • 5.7 million client connections •
5700 FTE
None
None
The problem • Fixed budget • How to optimally distribute
budget over departments • Balance between long term and short term goals
Amsterdam • Go fully electric by 2030 • 900.000 people
• 230.000 private cars • 10 years • 2610 working days • 1000 charging stations a day?
Our solution • Build agent based simulation of company •
Run millions of simulations up to 40 years in the future • Let Reinforcement Learning agent explore and optimize problem space
What do we need? • A physical model • A
bunch of agents manipulating this model
None
None
Reinforcement Learning
Reinforcement Learning st st+1 a
Deep-Q learning Simulation agent != Reinforcent learning Agent
State • Number of connection issues Reward • function of
• Budget spend • Average waiting time for new connection • Average Waiting time for fix of broken connection Action Decision of budget split
Project structure • 3 teams: • Data Engineers • Data
Scientists • Software Engineers
None
Evaluating 2 strategies • Compute in neo4j • Compute in
NumPy + NetworkIt
None
Rpy2 R has NA values in int arrays, numpy has
NaN in float arrays
Status report • Still working on simulations • Underestimated the
effort required to get this right • Maybe too ambitious? • Useful nonetheless
Research Engineer in Corporate environment • Different setting • Requires
different mindset • More budget • A lot of politics
pythonic.nl available again! • 1 June 2020 • First getting
married in Namibia • If we don’t get Corona banned
Questions?