Probabilistic forecasting of wind and waves for

offshore energy applications

Pierre Pinson, M. Fortin, P.-J. Trombe, H. Madsen

Technical University of Denmark, DTU Informaticsmail: pp@imm.dtu.dk - webpage: www.pierrepinson.com

Rave 2012 Conference - Session 8: ForecastingBremerhaven, 8-10 May 2012

Supported by the Danish PSO project Radar@Sea

1 / 11

Background: point forecasting

Forecasts (weather, energy production, etc.) are commonly provided in the form ofpoint forecasts (i.e. the conditional expectation for each look-ahead time)

5 10 15 20 25 30 35 40 450

10

20

30

40

50

60

70

80

90

100

look−ahead time [hours]

pow

er [%

of P

n]

pred.meas.

Different methodologies have been introduced and evaluated for wind, solar and wave power forecasting

Their level of accuracy is highly variable and depending upon external factors

2 / 11

Background: uncertainty ?!?

The level of accuracy is highly variable from one prediction to the other. This makesweather and energy-related forecasts somewhat hard to handle...

One cannot make optimal decisions from such predictions — an information ontheir (non-Gaussian) uncertainty is necessary!! (also accounting for correlationissues)

3 / 11

Forecasts for decision-making

Forecasts are to be used as input to (potentially complex) decision-making!

Actually, each forecast user has its own cost function(that he does not necessarily know...!!)

−1 −0.5 0 0.5 10

0.2

0.4

0.6

0.8

1

d

g(d)

Forecaster

−1 −0.5 0 0.5 10

2

4

6

8

10

12

d

g(d)

Trader (wind power only)

−1 −0.5 0 0.5 10

2

4

6

8

10

12

d

g(d)

Portfolio manager

So... What are the best forecasts for offshore energy-related applications??

4 / 11

1 Monitoring and short-term forecasting of weather/ocean and energy variables(0-6 hours)

2 Multivariate scenarios of wind and wave characteristics

5 / 11

Monitoring and short-term forecasts

X-band weather radars (range: 60kms) were installed at Horns Rev in 2009/2010 incollaboration with the Radar@Sea project

Supplemented by the DMI C-band radar (range: 240kms) at Rømø, they are usedfor

monitoring of weather conditions (safety)analysis of meteorological episodes (/climatology)improving short-term forecastsbetter control for episodes with intense fluctuations

6 / 11

Illustrative episode: 15-19.12.2010

Also see http://www.youtube.com/watch?v=YShQDCdVykM

7 / 11

Multivariate scenarios for wind and waves

Multivariate scenarios of wind and wave characteristics are a must-have as input todecision-making

Example work was performed based on ECMWF ensemble forecasts for FINO-1 over2010-2011

2010−01−01 2010−01−03 2010−01−05 2010−01−07 2010−01−09 2010−01−11 2010−01−13 2010−01−15

2

4

6

8

10

12

14

16

18

20

10 m

etre

win

d sp

eed

(m s

**−1

)

Ensemble membersEnsemble mean

wind speed

2010−01−01 2010−01−03 2010−01−05 2010−01−07 2010−01−09 2010−01−11 2010−01−13 2010−01−15

0.0

0.5

1.0

1.5

2.0

2.5

3.0

3.5

4.0

4.5

Sig

nific

ant w

ave

heig

ht (

m)

Ensemble membersEnsemble mean

significant wave height

These scenarios may beprovided by a number of public and commercial meteorological institutescalibrated for the purpose of optimal decision-makingused as input to complex decision-making problems related to offshore energy applications

8 / 11

Quality of such scenarios

ECMWF ensemble predictions already are extremely good at this kind of job...

lead time (h)

CR

PS

S (

m s

**−

1)

0 12 24 36 48 60 72 84 96 108 132 1560.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0 vs observationsvs analysis

wind speed

lead time (h)C

RP

SS

(m

)

0 12 24 36 48 60 72 84 96 108 132 1560.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0vs observationsvs analysis

significant wave height

and they can be significantly improved through appropriate multivariate calibrationprocedures

9 / 11

Outlook

Producing calibrated multivariate probabilistic forecasts of:input meteorological variables (wind and wave characteristics)power generation offshore (wind and wave energy)

Such forecasts also inform about space-time dependencies

Toy and real-world decision-making models should be set-up and analysed, for:optimal maintenance planningoptimal operation of wind-wave offshore energy plantetc.

Evaluating the quality and value of probabilistic forecasts (/scenarios)actual forecast quality(increased) value when used in decision-making

10 / 11

Thanks for your attention!

11 / 11