4.

0 Preliminaries of hydro-thermal scheduling (HTS)

4.1 Taxonomy of problem types

The goal of reservoir plant scheduling depends on the time frame of
interest.
1. Long-range (weeks to year): Here we need the load forecast and
expected water flow from a rainfall forecast. Then we can predict
the energy availability from the hydro facilities and compute the
necessary thermal energy.
2. Short-range (day to 1 week): Using results from (1) and more
precise load and water information, we can formulate a problem
where the solution yields the minimum cost of running the
thermal plants on an hour-to-hour basis.

In addition, there are 3 basic types of hydro scheduling problems.

1. All hydro system: Here we want to meet the demand with
minimum overall decrease in reservoir water level (stored
energy). For example, if you have two decoupled plants A and B,
and you run A all the time and never B, then plant B may spill.
The better approach is to run them both in such a way so that
neither plant spills. There are not many all-hydro systems in the
world, Nepal is one. But there are several that are close.
2. Mostly hydro system: New Zealand is such a system with 60%
hydro, 32% thermal, and the other 8% from various sources
(geothermal, wind, and biomass) [9]. Norway and Sweden
comprise another mostly hydro system with 29GW out of 30.5
GW being hydro. One of the best examples of a mostly hydro
system is Hydro Quebec. The Hydro Quebec region is shown in
Fig. 15b below together with all hydro facilities in the US.

20
 The below scale is
has been zoomed
out so as to read it.
The largest bubble
(2000MW & more)
corresponds to the
largest bubbles on
the map.

Fig 15b
This system uses hydro for producing over 97% of its energy, as
indicated in Fig. 15c below.

Fig. 15c
21
3. Mostly thermal system: Most systems with hydro qualify as this
type. PG&E in California is an example, as indicated in Fig. 15d.

Fig. 15d

Both (2) and (3) represent hydro-thermal coordination problems.

The table below provides a breakdown of several characteristics that

we have discussed – a taxonomy of problem types (note that no
coupling is intended between the different columns). We will study
a problem characterized by the bolded categories in the table below.
Plant Time frame Coupling Problem types
characteristics
Run of river Long range Coupled All Hydro
Pumped Short-range Decoupled Mostly hydro
storage
Reservoir Mostly
thermal

4.2 Terminology
Some basic terminology is illustrated in Fig. 16a and defined below.

22
 SPILLAGE

FOREBAY

INTAKE POWERHOUSE

PENSTOCK AFTERBAY

Fig. 16a
 Forebay: A lake or water impoundment (reservoir) before the
entrance to the power plant.
 Afterbay: A lake or water impoundment downstream from the
power plant that receives the water after it has passed through the
turbines.
 Penstock: The pipe leading from the water intake to the turbine.
 Intake: The entrance from the forebay to the penstock.
 Spillage: releasing water over the dam rather than through the
penstock. Some dams have spillways, as shown in Fig. 17, which
allow smolts (adolescent salmon) to pass without transiting
through the turbines.
 Fishladders: stepped runways that allow salmon to migrate
upstream, as shown in Fig. 16b (left). Fig 16b (right) also shows
the salmon’s amazing jumping ability.

Fig. 16b

23
 Fig. 17

4.0 The mostly-hydro case (Section 7.3.2 in book)

We assume in our development of this problem that there is one
thermal plant and one hydro plant. The one thermal plant is not a
restriction since we can always form a composite cost function
FS(PS) for our thermal plants. The one hydro plant represents a
simplification that we will eliminate later.

In the mostly-hydro case, the total rated hydro power is assumed to

be enough to meet the demand at any one period, but…the total
hydro energy is not enough to meet the demand over all periods, and
as a result, we must use the thermal generation.

We define the following nomenclature:

24
“Available water volume for use:” We can think of EH as a computed volume of water that will
not draw down the reservoir below a certain level, accounting for the upstream inflow.

 EH: total hydro energy available. It is proportional to the head

(gives distance water falls) and the volume of water above the
intake (gives mass of water to fall) or designated “available water
volume for use” in the time period of interest. Recall from (1):
EH  mgh (1)
And because m=Vρ, where V is water volume and ρ is the water
density (about 1000kg/m3 at normal temperatures), we have that
EH  Vgh (6)
We desire to use all of EH in a certain time period TMAX, where
j MAX
TMAX  nj (7)
j 1
Therefore
j MAX
EH   PHjn j (8)
j 1
where PHj is the hydro generation level at time period j.

 ES: total thermal energy used. Then

j MAX
ES   PSjn j
j 1 (9)

where PSj is the thermal generation level at time period j.

We will also assume that, in order to avoid start-up and shut-

down costs, we will utilize the thermal generation in only one
continuous period, that is, we will run the thermal generation for
the first NS periods and then shut it down, so that total thermal
run-time is:
25
 NS
TS  nj (10)
j 1
At this point, we do not know NS; therefore we do not know TS.

With the last assumption and the definition of NS, we may

express (9) as:
NS
ES   PSjn j
(11)
j 1
Equations (9) and (11) are equivalent, i.e.,
jMAX NS
ES  P n  P n
j 1
Sj j
j 1
Sj j

because, under the last assumption, the thermal generation levels

PSj, j=NS+1,…,jMAX, will be zero.
 EL: Total energy used by the load.
j MAX
EL   PLj n j
j 1 (12)

where PLj is the load during time interval j.

Observe that ES is a known quantity since

 we know PLj for all j which gives us EL, by (12);
 if we know the head and the water volume, then we know
EH from (6); and
ES  E L  E H (13)

26
 Our problem is then to minimize the cost of the thermal
generation subject to the requirement that we must supply an
amount of thermal energy equal to ES, i.e.,
NS
min FT   FS ( PSj )n j
j 1
subject to :
NS
ES   PSj n j  0 (14)

j 1
The Lagrangian is then
L ( PS1 ,...PSN S ,  )
NS  NS 
  FS ( PSj )n j    ES   PSj n j  0 (15)
j 1  j 1 
where α is the Lagrange multiplier on the energy constraint.

Applying the first-order optimality conditions, we obtain:

L F ( PSj )
 nj  n j  0, j  1,..., N S
PSj PSj (16)

27
 NS
L
 ES   PSj n j  0
 j 1 (17)

From (16), we observe that

F ( PSj )
 , j  1,..., N S
PSj (18)
Equation (18) implies that the thermal unit must be run at the same
generation level for the entire duration it is on-line, TS. Call this
power generation level PS*. We need to find this value. W&W
provide Fig.7.4 in the book, copied below, to illustrate the situation.

To do so, first observe that, with PSj=PS* for all j=1,…,NS, where
NS
TS   n j
(10)
j 1
then (17) becomes

28
 NS
L
 ES  PS*  n j  ES  PS*TS  0
 j 1 (19)

so that
ES
ES  PS*TS  TS 
PS* (20)
Let’s now assume that the thermal plant has a cost-rate function
given by

F ( PS )  A  BPS  CPS2 (21)

Since (18) requires the generation level to be the same throughout

 
the duration TS, the total cost of running the thermal plant is:

FT  F ( PS* )TS  A  BPS*  C ( PS* ) 2 TS (22)

Now substitute (20) into (22) to obtain

FT  F ( PS* )TS  A  BPS*  C ( PS* ) 2 P ES

*
S


 
 E S  A PS*
1
 B  CPS* 

(23)

Equation (23) is of value because, unlike (22) which is a function in

two unknowns (PS* and TS), (23) is a function in only a single
unkown, PS*, and it incorporates the equality constraint (19). So we
have transformed an equality-constrained optimization (14) to an
unconstrained optimization.

We can solve the unconstrained optimization using basic calculus,

which requires that the minimum of FT to occur when:

29
 FT
PS
*
 E  A P *
S

S   2
 C   0
 (24)
Manipulating (24) results in
A
C
 
* 2
PS (25)

Solving for PS* results in

A
PS*   (26)
C
Clearly we must have a positive value of generation, therefore
A
PS*  (27)
C
And we have solved our problem.
However, we should consider our solution – what does it mean?

To gain some insight into what (27) is telling us, recall that a
thermal plant cost-rate function F is related to its fuel-rate function
R (units of MBTU/hr) via:
F ( PS )  KRPS  (28)
where K is the fuel cost in $/MBTU. The fuel-rate function is then

RPS  
F ( PS )
K (29)
Then the (average, not incremental) heat rate of the plant is given by
R( PS ) F ( PS )
H PS   
PS KPS (30)

30
ASIDE (Potential source of confusion): In W&W, both in Chapter 2
(see pg 8) and in Chapter 7 (see pg 217), they use H for fuel rate and
therefore H/P is their heat rate, and they use f for fuel cost. In the
above equations, and in what follows, I am sticking to the same
notation that I used in my Cost-Curve notes earlier in the semester.

Assuming a quadratic form for F(PS), as we did in (21) above, (30)

becomes

A  BPS  CPS2
H PS  
KPS (31)
or

H PS  
1
K

APS 1  B  C PS   (32)
Differentiating, we obtain
dH PS  1
dPS

  APS  2  C  0
K
 (33)
Solving (33) for PS results in

A  CPS2 (34)
which results in
A
PS  
C (35)
Again, we require positive value of generation, therefore
A
PS  
C (36)
which is the same result obtained when minimizing cost rate.
Note the second derivative of (32) is
31
 d 2 H PS 
dPS2

1
K

2 APS  3  (37)
and this expression must be positive for positive values of P S,
therefore the extreme point found by (36) must be a minimum, that
is, we found the value of generation that minimizes the heat rate
function for the thermal plant.

Now you may recall from our Cost-Curve notes that Heat Rate, with
units of MBTU/MWHr, is proportional to the inverse of efficiency,
that is

H PS  
1
 (38)
So if our solution (36) minimizes heat rate, it maximizes efficiency.
This result provides us with the ability to understand the solution to
the hydro-thermal coordination problem that we have posed in this
section, for which we arrived at the same solution.

The implication is that by (13), repeated here for convenience,

ES  E L  E H (13)
we know the energy required by the thermal plant, our optimal
solution is to run the thermal plant at its point of maximum
efficiency for as long as it takes to produce this energy level ES.

Now that we have found this solution, it is hardly a surprise that it

should be this way, since operating at maximum efficiency is the
least-fuel usage way of supplying a fixed amount of energy.

Once we know PS*, then TS=ES/PS*, and then

32
  k 
N S  min k :  n j  TS 
 j 1 
which reads, “NS equals the minimum value of k such that the sum
of the durations from 1 to k is greater than or equal to TS.”

The hydro levels are then found from

PLEASE WORK THROUGH
PHj  PLj  PS* , j  1,..., N S EXAMPLE 7A IN THE TEXT.

PHj  PLj , j  N S  1,..., j MAX

(39)
5.0 The mostly-thermal case (Section 7.4 in book)

We will again assume one hydro generator and one thermal plant.

Our basic assumption for this problem is that we have a specified

amount of water energy we want to use over a given time, but we
never have enough hydro power to supply the entire load. Therefore
we must use some thermal generation at all times.

Let’s first consider the hydro input/output curve.

We have previously discussed (see notes on cost-curves) the

input/output (I/O) characteristic for a thermal plant, where we
plotted fuel input in MBTU/hr as a function of the power output Pg
in MW. The derivative of the I/O characteristic gives the
incremental fuel rate, in MBTU/MW-hr1.

1
When the I/O curve for a thermal plant is multiplied by the fuel cost in $/MBTU, we
obtain the cost rate curve in $/hr vs. MW. The derivative of the cost rate curve gives the
incremental cost rate, in $/MW-hr vs. MW.
33
We can do a similar thing for a hydro plant. In regards to the I/O
characteristic, the “fuel input” is given as qT, in volume of water/hr.
In the US, volume of water is often measured in Acre-ft (an Acre is
a unit of area equal to 4,046.87m2 or 43,560 ft2). Figure 18a
illustrates a representation of a hydroelectric plant I/O curve, which
shows that the power output is directly proportional to the water
flow rate up to the power rating of the plant, beyond which it
increases steeply due to increased water friction losses at the higher
flows.
q, (acre-ft/hr)

PMIN PRATED PMAX

Fig. 18a
Figure 18b plots the incremental flow rate dq/dt as a function of P.

34
 dq/dt, (acre-ft/MW-hr)

PMIN PRATED PMAX

Fig. 18b
We let qj be the water flow in acre-ft/hr for time interval nj. Then
jMAX
qTOT  n jq j (40)
j 1
where qTOT is the total water volume available for the hydro
generation in units of acre-feet.

Let’s also assume we have an I/O curve given by

q  q( PH ) (41)
and that it has the following form:
q  q( PH )  a  bPH (42)
We can now form our problem.

35
 NS
min FT   FS ( PSj )n j
j 1
subject to :
jMAX
 n j q j ( PHj )  qTOT (43)
j 1
PLj  PHj  PSj  0, j  1,..., j MAX
Recall the fuel scheduling problem (see (5) of FS notes), which was
jmax N
min  n j  Fij ( Pij )
j 1 i 1
subject to
N
PRj   Pij  PTj  0, j  1,..., j max
i 1
jmax
 n j qTj  qTOT
j 1
This is the exact same problem except in hydro-thermal scheduling
(HTS) we use composite thermal plant representation, whereas in
the above stated fuel scheduling (FS) problem, we used individual
thermal plant representation.

The Lagrangian of our HTS problem is

36
 L ( PS1 ,..., PSj max , 1 ,...,  j max ,  )

  
NS
  FS ( PSj )n j   j PLj  PHj  PSj
j 1
(44)
 jMAX 
   n j q j ( PHj )  qTOT 

 j 1 
 
And applying first-order conditions, we find for each interval that
q( PHk ) FSk ( PSk )
k   nk  nk
PHk PSk (45)
As in the FS problem, we have

FSk ( PSk )  A  BPSk  CPSk

2
(46)
and upon differentiating, we obtain
FSk ( PSk )
 B  2CPSk
PSk (47)
Substitution into (45) results in

k  nk
FSk ( PSk )
PSk
 nk B  2CPSk   (48)
from which we obtain
k  nk B
PSk  (49)
2Cnk
Equation (49) is the same as that which we obtained in the FS
problem (see (18) of FS notes).

37
There is a small difference in the equation for q (input/output curve,
MBTU/hr for FS or Acre-ft/hr for HTS), however, because in the
fuel scheduling problem, the energy constrained unit was thermal,
and therefore q was quadratic. Here, we use a linear expression for
q, per (42), repeated here for convenience:
q  q( PH )  a  bPH (42)
Differentiating, we obtain (compare to (22) of FS notes)
q
b
PH (50)
Substitution into (45) results in
q( PHk )
k   nk   nk b
PHk (51)
from which we obtain
k
  (52)
nk b
Substituting (48) into (52) we obtain

 

nk B  2CPSk   B  2CPSk
nk b b (53)
Solving (53) for PSK results in
 bB
PSk  (54)
2C
Equation (54) indicates that the thermal generation should be
constant across all time periods k=1,…,jmax, The only thing we need
to choose is γ. How do we choose γ?

38
W&W, pg. 220, states, “This is solved using the same techniques as
in Chapter 6.” This is true, but in this case, those techniques can be
considerably simplified because the “fuel” (water) is, in this case, a
linear function of the corresponding generation PH per (42). Because
of this, the incremental flow rate of hydro generation is a constant,
independent of the hydro generation level PH, as indicated in (50).
This means that the optimality condition of (45) does not directly
impose any requirement on PH (as it does PS); i.e., PH does not
appear in (45) since the differentiation with respect to PH is a
constant. This differs from the fuel-constrained problem where we
found that the optimality condition did impose a requirement on P T.
See equations (23) and (24) of the FS notes, which are repeated
below:
L
PTk

 k   nk bT  2cT PTk  0 
   nk bT
PTk  k
2cT  nk
We will describe the simplified procedure in what follows.
However, you should be clear that the simplification arises not
because of some inherent difference between the hydro scheduling
problem and the FS problem but rather because we here represent
the fuel I/O curve as linear instead of quadratic.

As in the fuel scheduling problem, γ may be thought of as a fuel

(water in this case) price, if we raise it, the supply of the fuel will
decrease; if we lower it, the supply of fuel will increase.

One can see this directly by substituting (54) into the power balance
equation
PLj  PHj  PSj  0, j  1,..., j MAX
which results in

39
  bB
PLj  PHj   0, j  1,..., j MAX
2C
Solving for PHj gives
 bB
PHj  PLj  , j  1,..., j MAX (55)
2C
Substituting (55) into (42) results in
  bB
q j  q( PHj )  a  b PLj  
 2C  (56)
Now it must be the case that the total water usage is qTOT according
to (40), repeated here for convenience:
jMAX
qTOT  n jq j (40)
j 1
Therefore (56) and (40) offer a simple way to solve the problem:
1. Guess γ
2. Use (56) to compute qj for all j=1,…,jmax
3. Compute
jMAX
Q n jq j
j 1
4. If | ΔqTOT|=|qTOT-Q|<ε, stop.
5. Else, if ΔqTOT>0 (qTOT>Q), decrease γ. If ΔqTOT<0 (qTOT<Q),
increase γ.
6. Go to 2.
Step 5 requires an expression for how much to increase/decrease γ.
This can be found in a similar way as we did for the FS problem, by
substituting (56) into (40) and then differentiating ∂qTOT/∂γ,
approximating as ΔqTOT/Δγ, and solving for Δγ. The result is
 qTOT  qTOT
   2 Observe in the update equation that
jmax
n jb2 b Tmax
j 1 2c
2c
positive ΔqTOT (we have not used
enough water) results in a price
decrease; negative ΔqTOT (we have
40 used too much water) results in a
price increase.
 j MAX

where TMAX  nj

j 1
In comparison to the FS solution approach (see Fig. 2 of FS notes),
this problem is simpler because it eliminates the need to iterate to
get PS and PH, which in the FS problem is required by the optimality
conditions of (45).

Mostly-thermal case with losses (pp 220-222)

W&W also consider representing losses in the problem of the

mostly-thermal case. In contrast to (43), the lossless problem
NS
min FT   FS ( PSj )n j
j 1
subject to :
jMAX (43)
 n j q j ( PHj )  qTOT
j 1
PLj  PHj  PSj  0, j  1,..., j MAX
our problem now becomes
NS
min FT   FS ( PSj )n j
j 1
subject to :
jMAX (57)
 n j q j ( PHj )  qTOT
j 1
PLj  P loss, j  PHj  PSj  0, j  1,..., j MAX
The Lagrangian becomes

41
 L ( PS1 ,..., PSj max , 1 ,...,  j max ,  )

  
NS
  FS ( PSj )n j   j PLj  P loss, j  PHj  PSj
j 1
(58)
 jMAX 
   n j q j ( PHj )  qTOT 

 j 1 
 
And applying first-order conditions, we find for each interval k that
L FSk ( PSk )  Ploss,k 
 nk 
 k   1  0
PSk PSk  PSk 
(59a)

L  Ploss,k  q ( P )
 k 
  1   nk k Hk  0
PHk  PHk  PHk (59b)
Solving the above for λk and rearranging slightly results in
FSk ( PSk ) Ploss,k
nk  k  k
PSk PSk (60a)

q ( P ) P
 nk k Hk  k loss,k  k
PHk PHk (60b)
which are identical to (7.28) and (7.29) in the text, which W&W
calls the coordination equations. In this case, the loss function Ploss,k
in (60b) will normally depend on PHk, and so (60b) will contain PHk.
Therefore we will need to use lambda-iteration to solve (60a) and
(60b). Figure 7.7 in W&W provides a general algorithm for solving
the overall HTS problem with losses, copied below.

42
 Note that there are three iterative loops:
 First (inner) loop: Lambda iteration solves
the coordination equations for a given time
interval. Actually, there is another loop here
which would, after computing lambda (and
corresponding generation levels) for a given
value of losses, would recompute losses and
iterate again.
 Second loop: Step through all of the time
intervals j=1,…,jmax.
 Third (outer) loop: Gamma adjustment to use
the desired amount of water.

REVIEW EXAMPLE 7B IN TEXT.

We now consider a series of increasingly complex HTS problems

for the mostly thermal case. We utilize a gradient approach for the
solution procedures.

Problem 1: This problem is the same as what we have addressed

above (mostly thermal case without losses), but we will utilize
slightly different notation, and we will be more explicit about our
assumptions, that is:
 Ignore inflow
 Starting volume=V0
 Ending volume=Vjmax
 Time intervals duration nj, j=1,…,jmax.
The presence of the starting and ending volume specifies a total
water volume usage which we will here denote as QTOT=V0-Vjmax.

Solution of this problem using the gradient search approach is given

below:

43
Step 1: Obtain starting solution as
QTOT V0  V j max
q k  q avg  jMAX
 jMAX
n j n j (61)
j 1 j 1

Step 2: Obtain the hydro generation level from qHk for each time
interval k=1,…,jmax, using (this just comes from the linear water rate
I/O equation (42)):
qk  a
PHk  (62)
b
Also obtain the thermal generation levels from:
PSk  PLk  PHk (63)

Step 3: Compute γk for all intervals k=1,…, jmax using:

B  2CPSk
k  (64)
b
which is the same as (53). If all γk are within ε of each other, then
the algorithm stops.

Step 4: Identify maximum and minimum γk (denoted qj+ and qj-)

from the computations made in Step 3, and then compute the change
in the total water released according to:
Q    j  j Q0 (65)
where ∆Q0 is chosen to be relatively small. Note that this relation
was used in the fuel scheduling notes as well. Justification was
given there as below:
The flow chart step “∆q=(γj+-γj-)∆q0 WHERE ∆q0 IS A CHOSEN SMALL
STEP” is not dimensionally correct as it stands, because gamma
has units of $/RE, and when multiplied by RE, gives $, consistent
with the above discussion regarding (43). You can assume,
44
 however, that the relation is really ∆q=[(γj+-γj-)/1][∆q0], where the “1”
has the same units as γ. Then we observe that if Δq0 has units of
RE, then so will Δq. Basically, this relation is just telling us that
if we want to correct two intervals j- and j+ for their fuel (or
water) usage, we should choose an amount of water to shift that is
proportional to the difference between the two interval’s gamma
values.
Step 5: Correct the water flow for intervals j+ and j- according to
Q
q j  q j 
n j (66)

Q
q j  q j 
n j (67)
Once this is done, we return to step 2.

Problem 2: This is the same as Problem 1, except we will assume a

rate of inflow to the reservoir of rk during interval k.

The only change in our previous solution strategy is in Step 1, where

we now need to include the inflow into our starting solution
according to
V0  V j max
q k  q avg  jMAX
 rk
n j (68)
j 1
This makes sense because an inflow is simply going to increase the
available outflow by that same amount.

Once this change is made, the procedure may proceed as indicated

in Problem 1.

45
Problem 3: Now assume that we have problem 2 except we will
account for limitations on the reservoir level of
Vmin  Vk  Vmax (69)
We must not violate this constraint in any time period.
To address this, we first must identify Vk, the water volume in
period k. This will be given by
Vk  Vk 1  Water Ink  Water Outk (70)
But the “water in” is the rate of inflow rk times the duration nk, and
the water out is the water flow rate qk times the duration nk, so that
(70) becomes:
Vk  Vk 1  rk nk  qk nk
Vk 1  rk  qk nk (71)
Equation (71) is referred to as the hydraulic continuity equation.

Solving (71) for qk results in

Vk 1  Vk
q k  rk  (72)
nk
Now recall that we have constraints on Vk as specified in (69).
Given a certain water level in the last period Vk-1,
 Letting Vk go to Vmin will result in the largest possible water flow
rate qk during interval k.
 Letting Vk go to Vmax will result in the least possible water flow
rate qk during interval k.
Therefore we may write that
qk min  qk  qk max (72)
where
Vk 1  Vmin
q k max  rk 
nk

46
 Vk 1  Vmax
q k min  rk 
nk
so that (72) becomes
Vk 1  Vmax V  Vmin
rk   q k  rk  k 1 (73)
nk nk
Note that, in addition to (72) (or (73)), there are also constraints on
water flow rate qk imposed the physical capabilities of the facilities
(water intake and penstock), i.e.,
qmin  qk  qmax (74)

Now what should be changed in our algorithm?

Recall Step 5 of our algorithm when we compute new values of qj+

and qj-. When considering constraints on water volume, we need to
check whether a new value of qj+ or qj- is in violation. If so, then the
water flow rate should be set to its limit. And so step 5 should read
as follows:
================================================
Step 5: Correct the water flow for intervals j+ and j- according to
Q
q j  q j 
n j (66)

Q
q j  q j 
n j (67)
If qj+>min(qmax, qkmax), then qj+= min(qmax, qkmax).
If qj-<max(qmin, qkmin), then qj-=max(qmin, qkmin).

Once this is done, we return to step 2.

47
================================================
The other change that is necessary is due to the fact that hitting a
constraint decouples the time intervals so that γ on one side of the
binding constraint will differ from γ on the other side of the binding
constraint.

You can think about what this means to our algorithm in two
different ways.
1. It makes no sense to try and draw the gamma values closer
together if they are chosen from separate sides of a binding
constraint, since there is no reason why they should be the
same. Therefore we need to choose the gamma values to draw
closer from the same side of the constraint.
2. Augmenting water release from two sides of a binding
constraint will necessarily result in violation of the constraint,
since the fact that the constraint is binding means it wants
more release in the direction of the violation. Augmenting
water release from only one side of the binding constraint will
necessarily result in no change to the flow rate at its limit,
since the net change in water release is zero, and since both
changes occur either before or after the binding constraint, it is
not possible that the bound flow rate should change.

Therefore the other change that should be made is in Step 4, where

we define the term “coherent” to include all time intervals for which
the gamma values should be the same, i.e., there should be no time
interval included within the chosen range for which a constraint is
binding.

Step 4 should now read as follows.

================================================
Step 4: For a given coherent set of time intervals, identify maximum
and minimum γk (denoted qj+ and qj-) from the computations made

48
in Step 3, and then compute the change in the total water released
according to:
Q    j  j Q0 (65)
where ∆Q0 is chosen to be relatively small.

================================================

[1] A. Mazer, “Electric power planning for regulated and deregulated markets,” 2007,
Wiley-Interscience.
[2] R. Kelin of Symbiotics, LLC, “ Presentation to NWPCC,” October 17. 2008, at
www.nwcouncil.org/energy/wind/meetings/2008/10/Default.htm
[3] R. Miller, “Wind Integration Utilizing Pumped Storage,” Norhwest Power and
Conservation Council, Oct. 17, 2008, available at
www.nwcouncil.org/energy/wind/meetings/2008/10/Default.htm.
[4] G. Tam of Electric Power Group, “Eagle Mountain Hydro-Electric Pumped Storage
project,” presented to Northwest Wind Integration Forum, Portland Oregon, Oct. 17, 2008,
available at www.nwcouncil.org/energy/wind/meetings/2008/10/Default.htm.
[5] M. Yeung, “Helms Pumped Storage Plant,” presented at the Northwest Wind
Integration Forum Workshop, Oct. 17, 2008, available at
www.nwcouncil.org/energy/wind/meetings/2008/10/Default.htm.
[6] Colorado River Commission of Nevada, “Laws of the Rivers: The Legal Regimes of
Major Interstate River Systems of The United States,” available on-line at
http://crc.nv.gov/docs/Laws_of_the_Rivers.pdf.
[7] www.nwd-wc.usace.army.mil/report/colmap.htm
[8] http://www.explainthatstuff.com/turbines.html
[9] I. Mason, S. Page and A. Williamson, “A 100% renewable electricity generation system
for New Zealand utilising hydro, wind, geothermal and biomass resources,” Energy Policy,
Volume 38, Issue 8, August 2010, Pages 3973-3984.

49