my presentation at icem 2017: from data mining to information extraction: using machine-learning to...

20
Climate Change Matteo De Felice (1) , L. Dubus (2) , A. Troccoli (3) (1) ENEA, Italy (2) EDF R&D, France (3) Univ. of East Anglia, UK From data mining to information extraction: using machine- learning to model hydro-power production in Europe

Upload: matteodefelice

Post on 22-Jan-2018

53 views

Category:

Science


0 download

TRANSCRIPT

Page 1: My presentation at ICEM 2017: From data mining to information extraction: using machine-learning to model hydro-power production in Europe

Climate Change

Matteo De Felice (1), L. Dubus (2), A.

Troccoli (3)

(1) ENEA, Italy

(2) EDF R&D, France

(3) Univ. of East Anglia, UK

From data mining to information

extraction: using machine-

learning to model hydro-power

production in Europe

Page 2: My presentation at ICEM 2017: From data mining to information extraction: using machine-learning to model hydro-power production in Europe

Climate

Change

R a t i o n a l e : E C E M ’ s G o a l s

European Climatic Energy Mixes (ECEM): A C3S Sectoral

Information Service for the Energy Sector

ECEM is a 27-month project to develop a proof-of-concept climate

service – or Demonstrator – for the European energy sector. It

will enable the energy industry and policy makers assess how well

different energy supply mixes in Europe will meet demand, over

different time horizons (from seasonal to long-term decadal

planning), focusing on the role climate has in the mixes.

Page 3: My presentation at ICEM 2017: From data mining to information extraction: using machine-learning to model hydro-power production in Europe

Climate

Change

• Hydropower is still the dominant RES in many countries: in EU-

28 in 2015 the largest RES accounting for 14.4% of total primary

energy production [source: EUROSTAT]

• It is a flexible source: it’s crucial in high RES penetration

scenarios

w h y ?

Page 4: My presentation at ICEM 2017: From data mining to information extraction: using machine-learning to model hydro-power production in Europe

Climate

Change We have three types of HP plants:

1. Run-of-river plants

2. Storage dams

3. Pumped storage

h y d r o p o w e r

Page 5: My presentation at ICEM 2017: From data mining to information extraction: using machine-learning to model hydro-power production in Europe

Climate

Change

I n s t a l l e d c a p a c i t y i n E u r o p e

Page 6: My presentation at ICEM 2017: From data mining to information extraction: using machine-learning to model hydro-power production in Europe

Climate

Change

• Modelling river flow from weather is hard: need to model

precipitation, run-off, evaporation, snow melt, etc.

• Water in reservoirs can be used for multiple purposes, e.g.

irrigation.

• There are other constraints, such as maintaining navigability on

rivers, fish ladders, water levels for recreation, water cooling for

thermal power plants, etc.

• We need information at basin-level!

C h a l l e n g e s

Page 7: My presentation at ICEM 2017: From data mining to information extraction: using machine-learning to model hydro-power production in Europe

Climate

Change • ENTSO-E data on hydro-power generation available on the Transparency Platform.

• Two types: 1) Installed capacity (Installed Generation Capacity Aggregated, 14.1.A) and 2) Hourly generation data (Aggregated Generation per Type, 16.1.B&C).

• Three typologies: pumped storage, run-of-river and poundage, and water reservoir.

• Data available since 1/1/2015 and for this work we have used two years of data (until 31/12/2016).

• Meteorological data (precipitation and snow cover) has been extracted from the ERA-INTERIM considering the period 2014-2016.

D a t a s o u r c e ( s )

Page 8: My presentation at ICEM 2017: From data mining to information extraction: using machine-learning to model hydro-power production in Europe

Climate

Change

R e s u l t s

Page 9: My presentation at ICEM 2017: From data mining to information extraction: using machine-learning to model hydro-power production in Europe

Climate

Change • Country-level averages

• Daily data

• Precipitation (ERA-INTERIM)

• Snow depth (ERA-INTERIM)

T h e p r e d i c t o r s

Page 10: My presentation at ICEM 2017: From data mining to information extraction: using machine-learning to model hydro-power production in Europe

Climate

Change

• Rolling sum of the last N days to predict generation at day t

• What is the best value of N?

• Our approach: find the lag that maximise the correlation

between the rolling sum and the generation

l a g g e d p r e d i c t o r s

Main question: is the connection

between meteorological

predictors and generation

instantaneous?

Page 11: My presentation at ICEM 2017: From data mining to information extraction: using machine-learning to model hydro-power production in Europe

Climate

Change

O p t i m i s i n g t h e l a g g e d p r e d i c t o r s

Page 12: My presentation at ICEM 2017: From data mining to information extraction: using machine-learning to model hydro-power production in Europe

Climate

Change

O p t i m i s i n g t h e l a g g e d p r e d i c t o r s

Page 13: My presentation at ICEM 2017: From data mining to information extraction: using machine-learning to model hydro-power production in Europe

Climate

Change

Page 14: My presentation at ICEM 2017: From data mining to information extraction: using machine-learning to model hydro-power production in Europe

Climate

Change

A p p r o a c h

ENTSO-E Transparency Platform Data

ERA-INTERIM Data

Page 15: My presentation at ICEM 2017: From data mining to information extraction: using machine-learning to model hydro-power production in Europe

Climate

Change

E x a m p l e s

Page 16: My presentation at ICEM 2017: From data mining to information extraction: using machine-learning to model hydro-power production in Europe

Climate

Change

R e s u l t s – F r a n c e , r u n - o f - r i v e r

Page 17: My presentation at ICEM 2017: From data mining to information extraction: using machine-learning to model hydro-power production in Europe

Climate

Change

R e s u l t s – F r a n c e , r u n - o f - r i v e r ( n o l a g )

Page 18: My presentation at ICEM 2017: From data mining to information extraction: using machine-learning to model hydro-power production in Europe

Climate

Change

R e s u l t s - S p a i n , r e s e r v o i r

Page 19: My presentation at ICEM 2017: From data mining to information extraction: using machine-learning to model hydro-power production in Europe

Climate

Change

R e s u l t s - S p a i n , r e s e r v o i r ( n o l a g )

Page 20: My presentation at ICEM 2017: From data mining to information extraction: using machine-learning to model hydro-power production in Europe

Climate

Change • Data-driven and generalised modelling of HP generation:

(surprisingly?) good results

• (Another) building block of a simulator of the European power

networks

• Climate-focused analysis on the historical data

• Possibility to use the model to perform predictions?

T a k e a w a y m e s s a g e s

http://ecem.climate.copernicus.eu/