Probabilistic forecasting of wind and waves for offshore ... · Background: point forecasting...
Transcript of Probabilistic forecasting of wind and waves for offshore ... · Background: point forecasting...
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: [email protected] - 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