28.02.2014 Views

A framework for joint management of regional water-energy ... - Orbit

A framework for joint management of regional water-energy ... - Orbit

A framework for joint management of regional water-energy ... - Orbit

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

The change in the objective function value at stage t with respect to a change<br />

in inflow during stage t – 1 is obtained from:<br />

*, k<br />

*, k k<br />

t<br />

F Q<br />

t<br />

tm ,<br />

= ⋅<br />

k<br />

t-1, m t, m t-1,<br />

m<br />

L<br />

k l,<br />

k l st<br />

( n)<br />

= ptm ,<br />

+ ltm ,<br />

g t+ 1 ⋅ç<br />

ç rt-<br />

1, t<br />

ç è<br />

l = 1 ÷ ø<br />

ç st<br />

-1( n)<br />

k<br />

= gtm<br />

,<br />

( n)<br />

F <br />

Q Q Q<br />

æ ö æ ö<br />

( n) ( n) ( n)<br />

å ÷ ç ÷<br />

(21)<br />

ç è ø<br />

*,k<br />

k<br />

Note that the partial derivative <strong>of</strong> F<br />

t<br />

with respect to Q<br />

tm ,<br />

is obtained by adding<br />

the impact <strong>of</strong> inflow changes on the current and the future operation <strong>of</strong><br />

the system. The partial derivative <strong>of</strong> current inflow with respect to previous<br />

inflow is obtained from the inflow model in (13). The vector <strong>of</strong> expected<br />

slopes with respect to inflows is given by:<br />

g<br />

K<br />

1 k<br />

tm ,<br />

n gtm<br />

,<br />

K<br />

k=<br />

1<br />

The constant scalar d is calculated as:<br />

tm ,<br />

( ) = å ( n)<br />

(22)<br />

1 K *, k T *<br />

tm ,<br />

= Ft + jtm ,<br />

Etm<br />

,<br />

K<br />

k = 1<br />

d<br />

å (23)<br />

A set <strong>of</strong> M cut parameters d tm ,<br />

, j , and<br />

tm ,<br />

g are generated at every stage t <strong>of</strong><br />

tm ,<br />

each backward iteration, and added to the set <strong>of</strong> L parameters (calculated in<br />

previous iterations) that approximate the FCF at stage t − 1. The number <strong>of</strong><br />

cuts L grows by M in each iteration, gradually improving the approximation<br />

until convergence.<br />

The creation <strong>of</strong> the backward openings and the cuts is illustrated in Figure 11.<br />

k<br />

The openings are created by adding the inflow noise vector Q<br />

tm ,<br />

to the corresponding<br />

storage E obtained from the mth <strong>for</strong>ward simulation (Figure 11a).<br />

tm ,<br />

*<br />

The K cuts created <strong>for</strong> each m <strong>for</strong>ward scenario (Figure 11b) are averaged to<br />

construct M cuts (Figure 11c). The maximum over the M cuts is the linear<br />

approximation to the FCF (Figure 11d).<br />

Note that the M resulting cuts will be added as constraints in all M scenarios<br />

at stage t – 1. In order to “share” the cuts among scenarios, the inflow scenario<br />

tree used in the <strong>for</strong>ward simulation must have common samples, i.e. the<br />

distribution <strong>of</strong> the inflow noise vector must be the same <strong>for</strong> all scenarios<br />

within a given stage t and independent <strong>of</strong> the inflow history. In this case, the<br />

30

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!