Stochastic Programming - Index of
Stochastic Programming - Index of
Stochastic Programming - Index of
You also want an ePaper? Increase the reach of your titles
YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.
150 STOCHASTIC PROGRAMMING<br />
2.8.2 Further developments<br />
The model shown above is very simplified. Modeling <strong>of</strong> real systems must take<br />
into account a number <strong>of</strong> other aspects as well. In this section, we list some<br />
<strong>of</strong> them to give you a feeling for what may happen.<br />
First, these models are traditionally set in a context where the major goal is<br />
to meet demand, rather than maximize pr<strong>of</strong>it. In a pure hydro based system,<br />
the goal is then to obtain as much energy as possible from the available water<br />
(which <strong>of</strong> course is still uncertain). In a system with other sources for energy<br />
as well, we also have to take into account the cost <strong>of</strong> these sources, for example<br />
natural gas, oil or nuclear power.<br />
Obviously, in a model as simple as ours, maximizing the amount <strong>of</strong> energy<br />
obtained from the available water resources makes little sense, as we have<br />
(implicitly) assumed that the amount <strong>of</strong> energy we get from 1m 3 <strong>of</strong> water<br />
is fixed. The reality is normally different. First, the turbines are not equally<br />
efficient at all production levels. They have some optimal (below maximal)<br />
production levels where the amount <strong>of</strong> energy per m 3 water is optimized.<br />
Generally, the function describing energy production as a result <strong>of</strong> water usage<br />
in a power plant with several turbines is neither convex nor monotone. In<br />
particular, the non-convexity is serious. It stems from physical properties <strong>of</strong><br />
the turbines.<br />
But there is more than that. The energy production also depends on the<br />
head (hydrostatic pressure) that applies at a station during a period. It is<br />
common to measure water pressure as the height <strong>of</strong> the water column having<br />
the given pressure at the bottom. This is particularly complicated if the<br />
water released from one power plant is submerged in the reservoir <strong>of</strong> the<br />
downstream power plant. In this case the head <strong>of</strong> the upper station will depend<br />
on the reservoir level <strong>of</strong> the lower station, generating another source <strong>of</strong> nonconvexities.<br />
Traditionally, these models have been solved using stochastic dynamic<br />
programming. This can work reasonably well as long as the dimension <strong>of</strong><br />
the state space is small. A requirement in stochastic dynamic programming<br />
is that there is independence between periods. Hence, if water inflow in one<br />
period (stage) is correlated to that <strong>of</strong> the previous period(s), the state space<br />
must be expanded to contain the inflow in these previous period(s). If this<br />
happens, SDP is soon out <strong>of</strong> business.<br />
Furthermore, in deregulated markets it may be necessary to include price<br />
as a random variable. Price is correlated to inflow in the present period, but<br />
even more to inflow in earlier periods through the reservoir levels. This creates<br />
dependencies which are very hard to tackle in SDP.<br />
Hence, researchers have turned to other methods, for example scenario<br />
aggregation, where dependencies are <strong>of</strong> no concern. So far, it is not clear<br />
how successful this will be.