wind power scenario generation geoffrey pritchard university of auckland by regression clustering
TRANSCRIPT
![Page 1: Wind power scenario generation Geoffrey Pritchard University of Auckland by regression clustering](https://reader036.vdocuments.us/reader036/viewer/2022070308/551c4b14550346a5458b4a3f/html5/thumbnails/1.jpg)
Wind power scenario generation
Geoffrey PritchardUniversity of Auckland
by regression clustering
![Page 2: Wind power scenario generation Geoffrey Pritchard University of Auckland by regression clustering](https://reader036.vdocuments.us/reader036/viewer/2022070308/551c4b14550346a5458b4a3f/html5/thumbnails/2.jpg)
Scenarios for stochastic optimization
• Uncertain problem data represented by a probability distribution.
• For computational tractability, need a finite discrete distribution, i.e. a collection of scenarios.
Make decision here
?
![Page 3: Wind power scenario generation Geoffrey Pritchard University of Auckland by regression clustering](https://reader036.vdocuments.us/reader036/viewer/2022070308/551c4b14550346a5458b4a3f/html5/thumbnails/3.jpg)
Power system applications
• Wind power generation, 2 hours from now.
• Inflow to hydroelectric reservoir, over the next week.
Typical problems solved repeatedly:
– Need a procedure to generate scenarios for many problem instances, not just one.
![Page 4: Wind power scenario generation Geoffrey Pritchard University of Auckland by regression clustering](https://reader036.vdocuments.us/reader036/viewer/2022070308/551c4b14550346a5458b4a3f/html5/thumbnails/4.jpg)
Situation-dependent uncertainty
• Scenarios represent the conditional distribution of the variable(s) of interest, given some known information x.
• Different problem instances have different x.
![Page 5: Wind power scenario generation Geoffrey Pritchard University of Auckland by regression clustering](https://reader036.vdocuments.us/reader036/viewer/2022070308/551c4b14550346a5458b4a3f/html5/thumbnails/5.jpg)
Change in wind power over next 2hrTararua/Te Apiti 28/5/2004-31/3/2010
![Page 6: Wind power scenario generation Geoffrey Pritchard University of Auckland by regression clustering](https://reader036.vdocuments.us/reader036/viewer/2022070308/551c4b14550346a5458b4a3f/html5/thumbnails/6.jpg)
Change in wind power over next 2hrTararua/Te Apiti 28/5/2004-31/3/2010
Change in wind power: 7 discrete scenarios
Each scenario is a function of the present wind power x.
![Page 7: Wind power scenario generation Geoffrey Pritchard University of Auckland by regression clustering](https://reader036.vdocuments.us/reader036/viewer/2022070308/551c4b14550346a5458b4a3f/html5/thumbnails/7.jpg)
Change in wind power over next 2hrTararua/Te Apiti 28/5/2004-31/3/2010
Change in wind power: 7 discrete scenarios
Each scenario is a function of the present wind power x.
![Page 8: Wind power scenario generation Geoffrey Pritchard University of Auckland by regression clustering](https://reader036.vdocuments.us/reader036/viewer/2022070308/551c4b14550346a5458b4a3f/html5/thumbnails/8.jpg)
• Have data xi and yi for i=1,…n
x
y
Scenarios by quantile regression
![Page 9: Wind power scenario generation Geoffrey Pritchard University of Auckland by regression clustering](https://reader036.vdocuments.us/reader036/viewer/2022070308/551c4b14550346a5458b4a3f/html5/thumbnails/9.jpg)
• Have data xi and yi for i=1,…n
• Want scenarios for y, given x.
x
y
Scenarios by quantile regression
![Page 10: Wind power scenario generation Geoffrey Pritchard University of Auckland by regression clustering](https://reader036.vdocuments.us/reader036/viewer/2022070308/551c4b14550346a5458b4a3f/html5/thumbnails/10.jpg)
Scenarios by quantile regression
• Have data xi and yi for i=1,…n
• Want scenarios for y, given x.
• Quantile regression: choose scenario sk() to
minimize i k( yi – sk(xi) )
for a suitable loss function k().
x
y
![Page 11: Wind power scenario generation Geoffrey Pritchard University of Auckland by regression clustering](https://reader036.vdocuments.us/reader036/viewer/2022070308/551c4b14550346a5458b4a3f/html5/thumbnails/11.jpg)
Quantile regression fitting
• For a scenario at quantile , is the loss function
![Page 12: Wind power scenario generation Geoffrey Pritchard University of Auckland by regression clustering](https://reader036.vdocuments.us/reader036/viewer/2022070308/551c4b14550346a5458b4a3f/html5/thumbnails/12.jpg)
Scenarios as functions
• Choose each scenario to be linear on a feature space:
sk(x) = j jkbj(x)
• Typically bj() are basis functions (e.g. cubic splines).
• The quantile regression problem is then a linear program.
![Page 13: Wind power scenario generation Geoffrey Pritchard University of Auckland by regression clustering](https://reader036.vdocuments.us/reader036/viewer/2022070308/551c4b14550346a5458b4a3f/html5/thumbnails/13.jpg)
Change in wind power over next 2hrTararua/Te Apiti 28/5/2004-31/3/2010
Change in wind power: 7 discrete scenarios
Equally likely scenarios, modelled by quantiles 1/14, 3/14, … 13/14.
![Page 14: Wind power scenario generation Geoffrey Pritchard University of Auckland by regression clustering](https://reader036.vdocuments.us/reader036/viewer/2022070308/551c4b14550346a5458b4a3f/html5/thumbnails/14.jpg)
Quantile regression: pros and cons
• Each scenario has its own model. Scenario models are fitted separately.
• Fitting is computationally easy.
• Scenarios have fixed probabilities. Events with low probability but high importance may be left out.
![Page 15: Wind power scenario generation Geoffrey Pritchard University of Auckland by regression clustering](https://reader036.vdocuments.us/reader036/viewer/2022070308/551c4b14550346a5458b4a3f/html5/thumbnails/15.jpg)
Another way to choose scenarios
… choose scenarios to minimize expected distance of a random point to the nearest scenario. (Wasserstein approximation.)
Robust to general stochastic optimization problems.
Given one probability distribution …
![Page 16: Wind power scenario generation Geoffrey Pritchard University of Auckland by regression clustering](https://reader036.vdocuments.us/reader036/viewer/2022070308/551c4b14550346a5458b4a3f/html5/thumbnails/16.jpg)
Scenarios for conditional distributions
• Have data xi and yi for i=1,…n
• Want scenarios for y, given x.
x
y
![Page 17: Wind power scenario generation Geoffrey Pritchard University of Auckland by regression clustering](https://reader036.vdocuments.us/reader036/viewer/2022070308/551c4b14550346a5458b4a3f/html5/thumbnails/17.jpg)
Scenarios for conditional distributions
• Have data xi and yi for i=1,…n
• Want scenarios for y, given x.
• Wasserstein:
minimize i mink | yi – sk(xi) |
over scenarios sk() chosen from some function space.
x
y
![Page 18: Wind power scenario generation Geoffrey Pritchard University of Auckland by regression clustering](https://reader036.vdocuments.us/reader036/viewer/2022070308/551c4b14550346a5458b4a3f/html5/thumbnails/18.jpg)
Scenarios as functions
• Choose each scenario to be linear on a feature space:
sk(x) = j jkbj(x)
• Typically bj() are basis functions (e.g. cubic splines).
• The Wasserstein approximation problem is then a MILP with SOS1 constraints (not that that helps).
![Page 19: Wind power scenario generation Geoffrey Pritchard University of Auckland by regression clustering](https://reader036.vdocuments.us/reader036/viewer/2022070308/551c4b14550346a5458b4a3f/html5/thumbnails/19.jpg)
Algorithm: clustering regression
Let each observation (xi,yi) be assigned to a scenario k(i). Choose alternately
• the functions sk
• the assignments k(i)
to minimize
i | yi – sk(i)(xi) |,
until convergence (cf. k-means clustering algorithm).
![Page 20: Wind power scenario generation Geoffrey Pritchard University of Auckland by regression clustering](https://reader036.vdocuments.us/reader036/viewer/2022070308/551c4b14550346a5458b4a3f/html5/thumbnails/20.jpg)
Clustering regression
Let each observation (xi,yi) be assigned to a scenario k(i). Choose alternately
• the functions sk
• the assignments k(i)
to minimize
i | yi – sk(i)(xi) |,
until convergence (cf. k-means clustering algorithm).
For univariate y, a median regression problem
![Page 21: Wind power scenario generation Geoffrey Pritchard University of Auckland by regression clustering](https://reader036.vdocuments.us/reader036/viewer/2022070308/551c4b14550346a5458b4a3f/html5/thumbnails/21.jpg)
Example: wind powerExample: wind power, next 2 hours
![Page 22: Wind power scenario generation Geoffrey Pritchard University of Auckland by regression clustering](https://reader036.vdocuments.us/reader036/viewer/2022070308/551c4b14550346a5458b4a3f/html5/thumbnails/22.jpg)
![Page 23: Wind power scenario generation Geoffrey Pritchard University of Auckland by regression clustering](https://reader036.vdocuments.us/reader036/viewer/2022070308/551c4b14550346a5458b4a3f/html5/thumbnails/23.jpg)
![Page 24: Wind power scenario generation Geoffrey Pritchard University of Auckland by regression clustering](https://reader036.vdocuments.us/reader036/viewer/2022070308/551c4b14550346a5458b4a3f/html5/thumbnails/24.jpg)
![Page 25: Wind power scenario generation Geoffrey Pritchard University of Auckland by regression clustering](https://reader036.vdocuments.us/reader036/viewer/2022070308/551c4b14550346a5458b4a3f/html5/thumbnails/25.jpg)
![Page 26: Wind power scenario generation Geoffrey Pritchard University of Auckland by regression clustering](https://reader036.vdocuments.us/reader036/viewer/2022070308/551c4b14550346a5458b4a3f/html5/thumbnails/26.jpg)
Scenario probabilities
Each scenario gets a probability: that of the part of the distribution closest to it.
Given one probability distribution …
![Page 27: Wind power scenario generation Geoffrey Pritchard University of Auckland by regression clustering](https://reader036.vdocuments.us/reader036/viewer/2022070308/551c4b14550346a5458b4a3f/html5/thumbnails/27.jpg)
• Probability pk(x) of scenario k must reflect the local density of observations (xi , yi) near (x, sk(x)).
• Multinomial logistic regression: probabilities proportional to
exp(j jkbj(x))
where jk are to be found.
Conditional scenario probabilities
![Page 28: Wind power scenario generation Geoffrey Pritchard University of Auckland by regression clustering](https://reader036.vdocuments.us/reader036/viewer/2022070308/551c4b14550346a5458b4a3f/html5/thumbnails/28.jpg)
Wind: scenarios and probabilities
![Page 29: Wind power scenario generation Geoffrey Pritchard University of Auckland by regression clustering](https://reader036.vdocuments.us/reader036/viewer/2022070308/551c4b14550346a5458b4a3f/html5/thumbnails/29.jpg)
9%
7%
3%
90%
33%
70% 41%
26%
21%
Wind: scenarios and probabilities
![Page 30: Wind power scenario generation Geoffrey Pritchard University of Auckland by regression clustering](https://reader036.vdocuments.us/reader036/viewer/2022070308/551c4b14550346a5458b4a3f/html5/thumbnails/30.jpg)
The End
![Page 31: Wind power scenario generation Geoffrey Pritchard University of Auckland by regression clustering](https://reader036.vdocuments.us/reader036/viewer/2022070308/551c4b14550346a5458b4a3f/html5/thumbnails/31.jpg)
Wind power 2hr from now: lowest scenario,conditional on present power/wind direction
![Page 32: Wind power scenario generation Geoffrey Pritchard University of Auckland by regression clustering](https://reader036.vdocuments.us/reader036/viewer/2022070308/551c4b14550346a5458b4a3f/html5/thumbnails/32.jpg)
Wind power 2hr from now: lowest scenario,conditional on present power/wind direction