Title: Modelling inflows for water valuation
1Modelling inflows for water valuation
Dr. Geoffrey Pritchard University of Auckland /
EPOC
2Inflows where it all starts
CATCHMENTS
thermal generation
reservoirs
hydro generation
transmission
consumption
We want inflow scenarios for use with
generation/power system models - in a form
useful for optimization. Historical inflow
sequences work for back-testing of a known
strategy - but not for optimization (will be
clairvoyant).
3Hydro-thermal scheduling
- The problem Operate a combination of hydro and
thermal power stations - - meeting demand, etc.
- - at least cost (fuel, shortage).
- Assume a mechanism (wholesale market, or single
system operator) capable of solving this problem.
4SDDP for hydro-thermal scheduling
Week 7
Week 8
Week 6
5SDDP for hydro-thermal scheduling
Week 7
Week 8
Week 6
min (present cost) E future cost s.t.
(satisfy demand, etc.)
6SDDP for hydro-thermal scheduling
Week 7
Week 8
Week 6
min (present cost) E future cost s.t.
(satisfy demand, etc.)
S ps
s
The critical step requires estimating expected
cost (the expected future cost for earlier
stages) so uncertainty (from inflows) must be
modelled with discrete scenarios. - lognormal,
Pearson III, or other continuous distributions
wont do.
7(No Transcript)
8Inflow scenarios for a single week
(1/7/2014 7/7/2014, say)
The historical record gives one scenario per
historical year - may be too many scenarios, or
too few - historical extreme events can recur,
but only in the identical week of the year
9Scenarios by quantile regression
- Each scenario has its own curve.
- Any number of scenarios, possibly with unequal
probabilities. - - computationally intensive models
- Smooth seasonal variation.
- - model interpretation
10Serial dependence
Inflow scenarios for successive weeks should not
just be sampled independently.
11Serial dependence affects the distribution of
total inflow over periods longer than 1 week.
(Model simulated for 100 x 62 years, independent
weekly inflows.)
12Variance inflation
- Keep the assumed independence of weekly inflows,
but modify their distribution to increase its
variance. - Wrong in two ways, but hopefully the errors
cancel. - Used e.g. in SPECTRA.
13Variance inflation
- Keep the assumed independence of weekly inflows,
but modify their distribution to increase its
variance. - Wrong in two ways, but hopefully the errors
cancel. - Used e.g. in SPECTRA.
14With variance inflation, inflow distribution is
wrong over 1 week but not bad over 4 months.
(Model simulated for 100 x 62 years, independent
weekly inflows with vif2.2.)
15Explicit serial dependence
- Inflow is a random linear (or concave) function
of inflow (or a state variable) from previous
stage(s) - Xt Ft(Xt-1) (Ft
random, i.e. scenario-dependent) - Commonest type is log-autoregressive
- log Xt b log Xt-1 a et
(et random) - General linear form (ideal for SDDP)
- Xt At BtXt-1
(At, Bt random)
16Serially dependent models
62 scenarios derived from regression residuals
16 scenarios fitted by quantile regression
Models fitted to all data, shown for week
beginning 2013-08-31
17Serially dependent model can match inflow
distribution over 1-week and 4-month timescales.
(Model simulated for 100 x 62 years, dependent
weekly inflows, general linear form.)
18A test problem
- Challenging fictional system based on Waitaki
catchment inflows. - Storage capacity 1000 GWh (cf. real Waitaki
lakes 2800 GWh) - Generation capacity 1749 MW hydro, 900 MW
thermal - Demand 1550 MW, constant
- Thermal fuel 50 / MWh, VOLL 1000 / MWh
- Test problem a dry winter.
- 35 weeks (2 April 2 December)
- Initial storage 336 GWh
- Initial inflow 500 MW (56 of average)
- Solved with Doasa 2.0 (EPOCs SDDP code).
19Results optimal strategy
Inflow model Lost load (MW, probability) Spill (MW, probability) Energy price (/MWh)
dependent (general linear) 9.37, 28 2.90, 6 251
independent (vif) 8.71, 23 3.21, 12 220
independent (uncorrected) 1.59, 9 0.14, 1 112
(Quantities are expected averages over full time
horizon probabilities are for any shortage/spill
within time horizon.)
20(No Transcript)
21(No Transcript)
22(No Transcript)
23(No Transcript)
24(No Transcript)