You are on page 1of 7

1

A Mathematical Model for the Efficiency


Curves of Hydroelectric units
A. L. Diniz, Member, IEEE, P. P. I. Esteves, C. A. Sagastizbal

Abstract- Hydropower generation is a function of the


discharge of the generating units and the difference between the
forebay and tailrace levels of the reservoir, and is subject to
penstock head losses and to the generation unit efficiency factor.
This factor is in turn a function of the turbined flow and the head
in the reservoir, and is usually obtained from the hill curves - or
hill diagrams - of a unit.
In the mid / short-term operation planning of hydrothermal
systems, the unit efficiency factors are usually considered
constant. However, when a hydro producer aims at determining
an hourly dispatch according to a given daily or weekly target set
by the system operator, it is essential to consider variations in
each unit efficiency factor with respect to operating conditions, in
order to distribute the generation of the plant among units in a
more economic way.
This paper studies two mathematical models for the turbinegenerator efficiency. Model coefficients are computed by a linear
regression technique using points taken from the level sets of a
typical hill curve. For each model, different variants are
considered in order to satisfy some specific conditions.
By analyzing each models results, the variants that are most
suitable for the dispatch problem of a hydro utility are chosen.
For two important hydro plants in the Brazilian system, we give a
comparative assessment between the generation obtained when
considering an average unit efficiency factor and the generation
resulting from considering the variation of efficiency with respect
to the operating conditions of the unit.
Index Terms-- Hydroelectric power generation, linear
regression, power generation dispatch, energy conversion.

I. INTRODUCTION

he power generated by a hydro plant is a function of


the turbined flow and the water head in the reservoir, and
it is an important aspect to be considered in models for the
operation planning of hydrothermal power systems, see for
instance [1], [2]. The level of accuracy used to represent the
hydropower generation function depends on the problem
addressed. For example, in the long term planning, it is quite
usual to consider an average productivity (MW/(m3/s)) for
each plant, with correction factors to consider its variation
with the storage level of the reservoirs. In the mid and short
term centralized dispatch of predominantly hydro systems,

A. L. Diniz is with CEPEL, Electric Power Research Center, Rio de Janeiro,


Brazil (email: diniz@cepel.br).
P. P. I. Esteves and C. A. Sagastizbal are with IMPA, Pure and Applied
Mathematics Institute, Rio de Janeiro, Brazil (email: sagastiz@impa.br)

1-4244-1298-6/07/$25.00 2007 IEEE.

hydro power is represented by a nonlinear function of the


turbined flow and of the storage in the reservoirs [3], and
sometimes also of the spillage [4].
Usually, a constant turbine-generator efficiency is
considered in the literature. Nevertheless, this efficiency,
which is associated to converting the net water head potential
energy in the reservoirs into mechanical energy in the turbines
and later on into electrical energy in the generators, is a
function of the turbined flow and of the water head in the
reservoir. Efficiency factors are often described by the hill
curves - or hill diagrams - of the generating unit [5].
In the self scheduling problem of a hydro plant [6], or in
the hydro unit commitment subproblem [5], that originates
from decompositions of the short-term hydrothermal dispatch
problem, hydro plants are to be dispatched during the
following day for given energy prices or daily generation /
discharge targets. In these situations, each hydro unit is
usually considered individually, by taking into account not
only the hill curves, but also forbidden operation zones.
Forbidden zones correspond to those ranges of generation
where the unit cannot operate, due to cavitation and vibration
effects.
The objective of this paper is to build a mathematical
model for the hydro unit efficiency factor, by applying a
regression technique from points obtained in the hill chart
level sets of a typical hydro unit. We consider two
formulations. The first one is a 2nd degree polynomial
functional presented previously in [5], [7]. The second
formulation is derived from the first one by adding terms up
to the 4th degree. Variants of these two basic formulations are
proposed and compared so as to obtain a robust model related
to several possible operating conditions of the unit. A
simulation of use of the best alternatives in the hourly
dispatch of two hydro plants is performed, and the deviations
from the operation obtained by using a constant turbinegenerator efficiency are presented.
This paper is organized as follows. Section 2 describes the
hydro power generation function and makes a review of
different formulations to model the hill curves. Section 3
presents the expressions used in this work to model theses
curves, as well as some variants and a normalization
procedure proposed. In section 4 the phases in the
development of the model are detailed. Numerical results are
shown in section 5 and the final conclusions are given in
section 6.

II. HYDRO POWER GENERATION FUNCTION


The power (gh) of a hydro generating unit depends on the
turbined flow (q), on the design specifications of the unit represented by the turbine ( t ) and generation ( g )
efficiency factors - and on the net water head (h), as stated
below [8]:
gh = 9,81 10 3 t g q h ,
(1)
where the numerical factor makes the appropriate unit
conversions from (m) and (m3/s) to MW, taking into
consideration both the water density and the acceleration of
gravity.
The water head is in turn given by the expression:
h = hup (V ) hdw (Q, S ) hloss ,
(2)
where hup is the forebay level - which depends on the storage
of the reservoir (V) - and hdw is the tailrace level - which is a
function of the turbined flow of the whole plant (Q) and,
according to the plant site configuration, sometimes also of
the spillage (S). The last term in (2) accounts for penstock
head losses.
The turbine efficiency is a function of the net head and the
turbined flow:
t = t (h, q ) ,
(3)
while the generator efficiency depends only on the power [6]:
g = g ( gh ) .
(4)
The combined turbine-generator efficiency, referred to in
the sequel as unit efficiency, is a hill-shaped function of q
and h. Graphs of theses curves for different turbines can be
found, for instance, in [6], [9]-[11].
An example of the level sets of a typical hill curve is
shown in Fig. 1, adapted from [5]. The operation region of the
unit is restricted by the bold dashed lines, corresponding to
upper bounds on both the generator capacity and the turbined
flow.
340
320

q (m3/s)

120 MW

0.80

300

0.82
0.84
0.86

280
260

0.9
0.88 0

240

0.8
0 8 level sets for
turbinegenerator
efficiency

0.94
0.92

100

90

220

Level sets
for
Generator
Power

80 MW
0.78

200
180

Max. power
(generator)

Maximum discharge
(turbine)

0.

min. head
(32m)

70 MW

4 4 4
h (m)

max. head
(48m)

Fig. 1: Hill chart of a typical hydroelectric unit.

A. Production of the plant as a whole


The production function of the plant as a whole can be
obtained by summing up the outputs of all its units, assuming
there exists a certain order for the units to be started [10],
[12]. A graph of this aggregate function is shown in Fig. 2.

GH

Best efficiency points

3rd unit is switched on


2nd unit is switched on

Q
Fig. 2: Production function of a hydro plant with 3 units

This hydro plant input/output (I/O) relationship has been


modeled in various ways in the literature. Usual procedures
are to represent the power as a family of nonlinear concave
functions of the discharge, one at each given head [3], or as a
unique function of both discharge and storage [4]. It can be
noted by Fig. 2 that it is possible to define best efficiency
points for each number of switched on units. A special
attention can be given to these points while modeling the
function for the plant, as in [12].
In the works mentioned above, hydro generation is
obtained from the total turbined flow of the plant. As the
distribution of flow among the units is not considered, some
assumptions must be made for the unit efficiency factors. By
contrast, when the operation of each unit is considered
explicitly [5], [13], individual models for each unit efficiency
can be used.
B. Model for a single unit
As a hydro production is usually considered for the whole
plant in the short term operation of power systems, there are
not many works that model the efficiency function of
individual hydro units.
In [5], [7], a 2nd degree polynomial function of the
turbined flow and the water head is proposed for the turbinegenerator efficiency. It is a concave function, with a welldefined maximum that corresponds to the design point of the
unit. In [6], [11] the turbine and generator efficiency factors
are considered separately. Both works represent the generator
efficiency as a concave function of power. For the turbine
efficiency, the first work uses a hill curve of the turbined flow
and the head, while the latter uses several functions of the
turbined flow and the power, one for each value of head. In
[14], the efficiency of a Francis turbine is considered only as a
function of generation. It has been modeled as a polynomial
function with powers 1 and . In [15], the turbine efficiency
curve is presented relatively flat around the design flow.
III. MODELING THE EFFICIENCY FACTOR OF AN INDIVIDUAL
UNIT

We now introduce two basic models (A and B) for the


hydro unit efficiency. Model A is the one presented in [5] and
[7], while model B is proposed as a 4th degree polynomial
derived from model A by adding some terms, as shown in
equations (5a) and (5b) below:

Model A:
= a 00 + a10 h + a 01 q + a11 hq + a 20 h 2 + a 02 q 2

(5a)

Model B:

= b00 + b10 h + b01 q + b11 hq + b20 h 2 + b02 q 2 +

(5b)

+ b21 h 2 q + b12 hq 2 + b22 h 2 q 2

In these expressions, sub-indices of the coefficients


correspond to the degree of each term in variables h and q,
respectively, while the letter indicates the model to which the
coefficient belongs.
A. Normalization for h and q
We will see in section 4.3 that coefficients are computed
by solving a linear system. Since using absolute magnitudes
for h and q in (5a) and (5b) may result in an ill conditioned
linear system, we normalize the head and discharge values as
follows:
q q dp
~ h hdp
h=
, q~ =
,
(6)
hdp
q dp
where hdp and qdp are the head and discharge for the design
point of the turbine, respectively.
An additional advantage of using this normalization is that
coefficients of the normalized model do not depend on the
water head and discharge values at the design point, but rather
on the inclination of the lower and upper rays of the elliptical
level sets of the hill curve, and the magnitude of the
derivatives of the curve. Therefore, given a model for a
certain unit, identical models can be derived for other units
for study purposes, by keeping the same values of the
~
coefficients a~ij and bij in the normalized model. These
~
coefficients are obtained by using the numerical values of h
and q~ instead of h and q in equations (5a) and (5b). A further
linear transformation is performed to obtain the particular aij
and bij coefficients for each unit, using its head and discharge
values at the design point.
B. Variants of the basic models
Ideally, the best point to operate a hydro unit is at its
design point (hdp, qdp). For this reason, it may be desirable to
impose some conditions to the model at this particular point.
For example:
Condition I: maximum efficiency at the design point this requirement states that the partial derivatives of the model
in both h and q variables must be null. This leads to the
condition a10 = a01 = 0 for model A, and b10 = b01 = 0 for
model B.
Condition II: zero error at the design point - One may
desire the model to have a perfect match to the real curve at
the design point. With this assumption, additional conditions
a00 = dp for model A and b00 = dp for model B are imposed.
Specific conditions for some points: - although not
considered in our examples, such conditions may be useful if
the hydro utility wants to force the model to match the real

curve at some typical operating points for the unit, such as the
bound pairs (hmin, qdp), (hmax, qdp), (hdp, qmin), (hdp, qmax). In the
centralized dispatch, the system optimal operation not
necessarily leads to the operation of each unit at its optimum
point, as trade-offs may occur. Therefore, the plant generation
set by the system operator may force one or more of the units
of the plant not to operate near the design point, but rather at
the bound points mentioned above, depending on the system
conditions and constraints.
IV. PHASES OF THE MODELING
The construction of the proposed models and their variants
involves the following stages:
A. Data collection
First a set of triplets (hi,qi;i), i = 1, ... N, must be collected
from the real hill curve of a hydro unit. These points could be
obtained from unit specification tables or by real tests on the
turbine [16]. Due to lack of real data, we measured these
points in the hill chart shown in Fig. 1.
Two methods for measuring these points were applied. The
first method consists in measuring points (hi,qi) along level
sets for . In the second measurement method, points are
uniformly distributed along the graph, according to a
discretization grid of both head and discharge values, and the
efficiency factors are obtained through linear interpolation
of the values along the level sets.
B. Definitions of weights
To each measured point a weight i is assigned. This is
crucial to balance the distribution of the available data. Since
lower level sets are larger than higher level sets, often there
will be much more triplets (hi,qi;i), for lower values than for
higher values of i. Moreover, weights are useful to force the
model to be more accurate at some specific regions, such as
nearby the design point.
C. Calculation of coefficients by regression
Having the set of N triplets (hi,qi;i) obtained from the real
hill curve of the unit, each one with a given weight i, model
coefficients are computed by applying a linear regression
technique, i.e., by minimizing the average square error of the
efficiency values provided by the model, as compared to the
measured efficiency factors i.
Without loss of generality, suppose we must calculate the
coefficients for model A with no specifications about the
values of derivatives at the design point. If a denotes the
vector of coefficients [a00, a10, ... ,a02] and xi denotes the
vector with components [1, hi, qi, hiqi, hi2, qi2], we look for the
vector a that minimizes the weighted average square error
function E(a):

E (a) =

i =1

i ( i a T xi )

i =1

(7)

The optimal vector a* of model coefficients is then


obtained by solving the following 66 linear system [17]:

i xi xi T

i =1

[ a ] = x
N

(8)

i =1

which corresponds to the first order optimality conditions:


E *
j = 1,6,
(9)
(a ) = 0 ,
a j
where a1,,a6 are the components of a. Second order
optimality conditions are satisfied due to the convexity of
E(a).
Extending (8) to model B is straightforward. If some of the
additional conditions given in section III.B are also required,
some coefficients will be set a priori, and the resulting system
is reduced accordingly.

180
200
220
240
260
280
300
320
340

42,0
0,78
0,81
0,83
0,86
0,91
0,91
0,85
0,83

44,0
0,78
0,80
0,82
0,84
0,87
0,89
0,87
0,84

46,0
0,79
0,81
0,82
0,84
0,87
0,87
0,85

48,0
0,77
0,79
0,81
0,82
0,84
0,85
0,84

Different weights for the points were considered only for


the method of measurement along the level sets. The weights
adopted, shown in Table II, are related to the number of
points in each level set and the magnitude of the efficiency
values.
TABLE II
WEIGHTS CONSIDERED FOR THE POINTS MEASURED ALONG THE LEVEL SETS
0,94
0,92
0,90
0,88
0,86

V. NUMERICAL RESULTS

weight

72,0

12,0

6,0

4,5

In this section we present the results obtained for some


variants of models A and B. The database obtained by the two
measurement methods is shown in tables I-a and I-b.

0,84

0,82

0,80

0,78

weight

3,0

3,0

3,0

1,0

TABLE I-A
POINTS OBTAINED BY MEASURING ALONG THE LEVEL SETS
h(m) q(m3/s)
h(m) q(m3/s)
h(m) q(m3/s)
q(m3/s)

h(m)

= 0,94

41,56
=
40,90
41,60
42,00
42,40
42,00
41,20
=
38,90
40,00
42,00
42,70
44,00
44,40
44,30
44,00
42,00
41,50
40,00
39,10
=
37,40
38,00
40,00
40,50
42,00
43,70
44,00
45,40
45,50
44,00
42,00
41,00
40,00

265,8
0,92
260,0
260,0
262,8
272,1
273,0
269,5
0,90
248,7
244,3
254,7
260,0
271,3
280,0
285,0
286,6
283,3
280,0
269,5
260,0
0,88
238,2
234,7
237,3
240,0
248,7
260,0
262,6
280,0
291,6
295,8
286,6
280,0
273,9

38,40
38,00
37,40
=
35,90
36,00
38,00
40,00
41,80
42,00
44,00
44,70
46,00
46,60
46,80
46,00
44,00
43,20
42,00
40,00
38,00
37,30
36,00
35,90
=
34,80
36,00
38,00
40,00
42,00
43,80
44,00
46,00
46,20
48,00
48,80

260,0
256,5
240,0
0,86
230,4
228,7
223,4
229,5
240,0
241,7
254,7
260,0
272,1
280,0
300,0
304,3
303,4
300,0
293,3
280,0
266,0
260,0
244,3
240,0
0,84
220,0
213,6
211,8
220,0
229,5
240,0
240,8
258,2
260,0
280,0
300,0

48,00
47,60
46,00
44,00
43,60
42,00
40,20
40,00
38,00
37,70
36,00
35,50
34,10
=
33,10
34,00
35,70
36,00
38,00
39,10
40,00
42,00
44,00
44,10
46,00
46,50
48,00
48,00
46,00
44,00
42,00
40,50
40,00
38,00
37,90
36,00

318,1
320,0
324,3
321,7
320,0
312,5
300,0
298,3
282,5
280,0
265,2
260,0
240,0
0,82
220,0
211,8
200,0
198,3
197,5
200,0
202,6
209,0
219,1
220,0
234,7
240,0
260,0
335,6
340,0
337,1
328,7
320,0
316,5
300,8
300,0
284,1

35,60
34,00
33,60
32,40
=
32,00
32,30
33,00
34,00
44,00
42,00
36,00
38,00
40,00
42,00
43,70
44,00
46,00
46,90
48,00
41,10
40,00
38,00
37,50
36,00
34,90
34,00
33,20
32,00
=
42,90
44,00
46,00
47,00
48,00

280,0
265,2
260,0
240,0
0,80
206,0
200,0
193,3
190,0
350,9
345,0
186,6
186,6
188,3
193,3
200,0
201,7
213,9
220,0
229,5
340,0
334,5
323,6
320,0
310,0
300,0
290,0
280,0
260,0
0,78
180,0
183,4
193,3
200,0
210,0

TABLE I-B
POINTS OBTAINED BY MEASURING ALONG A DISCRETIZATION GRID
q (m3/s)
180
200
220
240
260
280
300
320
340

32,0
0,80
0,81
0,81
0,80
-

34,0
0,78
0,81
0,83
0,84
0,82
0,81
-

36,0
0,79
0,82
0,85
0,86
0,85
0,82
0,81
-

38,0
0,79
0,82
0,86
0,89
0,87
0,84
0,82
0,80
-

40,0
0,79
0,82
0,84
0,89
0,91
0,86
0,84
0,82
-

3,6

The identification and parameters for variants of models A


and B that were studied are given in Table III.
TABLE III
LIST OF VARIANTS FOR MODELS A AND B CONSIDERED IN THE STUDY
measurement
Condition
Variant ID
weights
Condition I
method
II
Grid
No
No
No
A9g
level sets
Yes
No
No
A9p
level sets
Yes
Yes
No
A7p
level sets
Yes
Yes
Yes
A6p
level sets
Yes
No
No
B6p
level sets
Yes
Yes
No
B4p
level sets
Yes
Yes
Yes
B3p

A. Model Results
Table IV shows the optimal coefficient vector a* for
models A9p, A7p, B6p and B3p, obtained by converting the
~ *, as described at the end of section
normalized vector a
III.A.
TABLE IV
OPTIMAL PARAMETERS FOR SOME VARIANTS STUDIED
A9p
A7p
B6p
B3p
a00 (b00)
a10 (b10)
a01 (b01)
a11 (b11)
a20 (b20)
a02 (b02)
a21 (b21)

+2,31006e+1

+1,96048e+1

-1,51495e+0

-2,60594e+0

-1,16890e+0

-9,71942e-1

+8,57224e-2

+7,01275e-3

-1,80742e-1

-1,53896e-1

+4,86940e-3

+1,25785e-1

+9,58249e-3

+8,06483e-3

+1,38907e-4

+1,53758e-4

+1,43625e-2

+1,16585e-2

-1,47865e-3

+2,52087e-5

+3,22799e-4

+2,73455e-4

-2,00394e-5

+2,00505e-3

-1,19052e-4

-9,81541e-5

a12 (b12)

-1,74129e-5

-1,46160e-5

a22 (b22)

+2,19181e-7

+1,80576e-7

The efficiency level sets for variants A9p and B3p are
shown in Fig. 3.

0 .7 0
0 .6 0

0 .7

350

0 .7 5
0 .8 4

0 .6 0

0 .7 0

0 .7 5

0 .7 8
0 .82

0 .8 4

0 .8 6

0 .8 8

0 .8 4

0 .8 8
0 .8 4

0 .7 8

0 .8 6

0 .8 0

150

20

0 .82

0 .6 0
0 .7 5

0 .7 8

0 .6 0

0 .70

0 .8 0

0 .7 5

200

0 .6 0

0 .8 2
0 .7 5

0 .8 0

250

0 .7 0

0 .7 8
0 .8 0

0 .8 6

300

0 .6 0

0 .7 8
0 .8 0
0 .8 2

0 .7 5

0 .7 0

25

30

35

40

45

50

55

60

65

0.82

400

0.90
0.86 0.88

0.80

0.60

0.84

0 .70

0.94
0.92
0.80

350

0.75
0.78

0.78
0.75

0.82

0.80
0.82

0.84

0.70
0.84
0.75

0.86

0.88

0.86

0 .80

0.8 0

0.86

0.82

0.84

0 .88

0.82

0.78

0.60
0.75
0.70

0.60
0.80

0.70
0.78
0.75

0.75
0.82
20

0.9
0.88
0.78

0.84

0.78

250

200

0.82

0.6 0

0.70

0.60
300

150

70

0.70

0.8 0.78
25

30

35

40

45

50

55

60

0.60
0.70 0.75
65

70

Fig. 3: Level sets for hill curves given by variants A9p (up) and B3p (down).

It can be noted from Fig 3 that model A looses its original


behavior for very small efficiency values, although this makes
no difference for use in an operation dispatch model, since
these regions lay beyond the domains of the feasible operating
region of the unit. In variant B3p, the condition of perfect
match at the design point was imposed, so the level curves of
this model attain a higher peak value than in model A9p (see
also Fig. 4).
B. Error analysis
To compare the models accuracy, unit efficiency values
were grouped in four intervals. The respective mean and
average absolute deviation as compared to the points in the
real hill curve are reported in Table V.
MEAN AND AVERAGE ABSOLUTE DEVIATIONS OF EACH VARIANT, GROUPED IN
DIFFERENT RANGES OF EFFICIENCY VALUES (% OF REAL EFFICIENCY VALUE)
A9g
A9p
A7p
A6p
(%)
abs.
abs.
abs.
abs.
mean
Mean
Mean
Mean
dev.
dev.
dev.
dev.
-3,47

3,47

-1,09

1,09

-1,12

1,12

+1,89

1,89

86 to 90

-0,42

0,99

+1,22

1,34

+1,24

1,34

+3,28

3,28

81 to 86

+0,73

0,88

+1,07

1,42

+1,11

1,46

+1,40

2,33

76 to 81

-0,58

0,81

-1,47

1,61

-1,45

1,63

-2,26

2,67

TOTAL

-0,12

1,04

+0,48

1,42

+0,50

1,44

+1,24

2,66

A9g

(%)

A9p

A7p

A6p

mean

abs.
dev.

> 90

-3,47

3,47

-1,09

1,09

-1,12

1,12

+1,89

1,89

86 to 90

-0,42

0,99

+1,22

1,34

+1,24

1,34

+3,28

3,28

81 to 86

+0,73

0,88

+1,07

1,42

+1,11

1,46

+1,40

2,33

76 to 81

-0,58

0,81

-1,47

1,61

-1,45

1,63

-2,26

2,67

TOTAL

-0,12

1,04

+0,48

1,42

+0,50

1,44

+1,24

2,66

Mean

abs.
dev.

0,94
0,92
0,90

hc

0,88

A6p

0,86

A7p

0,84

A9p

0,82
0,80

TABLE V

> 90

in general a hydro unit is operated near the higher efficiency


region, variant A9g, as well as any other with no points
weighting, is not adequate.
We see that the enforcement of perfect match for the
efficiency value and derivatives at the design point makes
variants A6p of model A and B3p of model B too optimistic
for the two higher efficiency regions (see also Fig. 4). In
addition, both mean and average absolute deviation for these
variants are big.
The remaining variants A9p, A7p, B6p, B4p show a better
fitting to efficiency values in the hill curve, with mean errors
around +0,50% and average absolute deviations below 1,60%.
Among these variants, we note a slight advantage for variants
A9p and A7p, both in overall and higher efficiency region
errors. The condition of zero derivatives at the design point
(variant A7p) does not seem to be much effective, because
this condition is already almost satisfied by variant A9p.
In Fig. 4a, comparison of efficiency values along the
highest efficiency path for variants A9p, A7p and A6p of
model A and the exact hill curve (hc) is given. The highest
efficiency path is the line in the h q plane where highest
efficiency is met for each value of head.

max. efficiency (p.u.)

400

Mean

abs.
dev.

Mean

abs.
dev.

From Table V we see that the minimum overall average


absolute deviation is given by model A9g. However, in this
model errors appear mostly in higher efficiency values (note
that model A9g assigns equal weights for all the points). Since

32,0

36,0

40,0

43,0

47,0

head ( m )

Fig. 4: Efficiency variation along the highest efficiency path for models A9p,
A7p and A6p, as compared to the real hill curve efficiency (hc).

We confirm in the graph the optimistic behavior of variant


A6p. The same occurs for the corresponding variant B3p of
model B. By contrast, variants A9p and A7p, the superior
ones in our analysis, are pessimistic near the highest
efficiency points and become more optimistic as efficiency
values decrease.
C. Simulation of the hourly operation of a hydro plant
In this analysis we check how generation for hydro units
varies when using average efficiency values in a short term
hydrothermal dispatch model developed at CEPEL Error!
Reference source not found., as compared to the generations
that would be obtained if exact hill curves efficiency factors
were considered instead.
As a first step, a dynamic economic dispatch with the
whole Brazilian system, consisting of 110 hydro plants, 4
pumped-storage units and 59 thermal units, was performed.
The study consists of 168 hourly time steps, with a cost-to-go
function at the end of the study given as an input by the
medium term model, DECOMP [1]. A unit efficiency value of

100
80

450

60

400

40

350
1

10

13

16

h (dp)
q (dp)

19

q (m3/s)

h (m )

500

120

450

110

400

100

350

90

gh (MW)

300
16

gh hc
ghhc

19

16

19

22

h (dp)

head

q (dp)

discharge

340

300
1

10

13

16

gh
ghmean
mean

19

ghhc
ghhc

22

Fig. 6: Comparison between consideration of an average unit efficiency and the


real hill curve efficiency Paulo Afonso Hydro Plant.

In Table VI we make a comparison, for these two plants,


between the generation obtained by using an average unit
efficiency (ghmean) and the generation obtained by using the
exact hill curve efficiency. The generation values during the
day are grouped by load level.

High

350

13

13

320

400

10

10

360

medium

380

450

gh mean
ghmean

400

500

300
1

Low

q (m3/s)

500

h (m )

130

TABLE VI
COMPARISON BETWEEN ghmean AND ghhc, GROUPED BY LOAD LEVEL.
Load level
Itumbiara
P. Afonso IV

22

head
discharge

the design point value.

gh (MW)

100% was set a priori for all hydro units.


In a second step, having the generation, discharge and net
head results given by the model, unit efficiency values for
each time interval and each unit were calculated by using
model A9p. The average efficiency during the week for each
unit was also computed. We made a comparison between:
generations obtained for each unit and each time interval
by applying the average weekly efficiency value of each
unit to the generations given by the model (this weekly
average efficiency would be considered in the dispatch
model in a feedback procedure), and
generations obtained for each unit and each time interval
by applying the corresponding efficiency value of model
A9p to the generations given by the model.
Results are detailed for two hydro plants: Itumbiara, a large
plant located at the southeast region, and Paulo Afonso IV, a
run-of-the-river plant at the San Francisco River basin. Due to
lack of real data for the hill curves of the corresponding units,
we adopted similar curves to the one shown in Fig. 1, with
appropriate adjustments to the feasible water head and
discharge ranges of each real unit. The main data for these
plants may be obtained at http://www.ons.org.br.

ghmean (MW)
ghhc (MW)
diff (%)
ghmean (MW)
ghhc (MW)
diff (%)
ghmean (MW)
ghhc (MW)
diff (%)

332,83
318,00
+4,67
363,09
364,41
-0,36
411,19
395,30
+4,02

317,81
319,60
-0,56
317,48
318,22
-0,23
329,42
322,03
+2,30

22

Fig. 5: Comparison between consideration of an average unit efficiency and the


real hill curve efficiency Itumbiara Hydro Plant.

The main results for the first day of study are reported in
Fig. 5 and 6. In the upper charts, head and discharge variation
along the day are shown, compared to the head and discharge
values at the design point. In the lower charts, we show the
generations obtained when considering an average efficiency
(ghmean) and the generations obtained when considering the
exact hill curve efficiency factor (ghhc). Water head variation
are very small during the day, but we note that, for Itumbiara
hydro plant, ghhc is lower than ghmean at the peak load level
hours, as discharge values distances from the design point
value. The opposite occurs for Paulo Afonso hydro plant: ghhc
is greater than ghmean because the discharge value is closer to

Differences between ghmean and ghhc are always within 5%.


These small differences justify the usual consideration in the
literature of constant average unit efficiency factors in the
centralized short term hydrothermal scheduling. Besides, the
representation of the hill curves of hydro units in this problem
may become very cumbersome in terms of computational
time.
On the other hand, in the self-scheduling problem of a
hydro utility, the consideration of exact unit efficiency factors
gives a more precise scheduling for the hydro plants, and
accumulation of small hourly energy savings may eventually
result in significant annual savings.
VI. CONCLUSIONS
This paper presented two mathematical models for the
combined turbine-generator efficiency of the hydroelectric
units. Some variants of both models were considered to

satisfy specific conditions at the design point of the unit.


Model parameters were calculated via linear regression using
points taken from the level sets of a typical hill curve.
The requirement of a perfect match to the real curve at the
design point of the unit results in a very optimistic model at
the higher efficiency points, with a greater average error along
the curve. In addition, although being able of controlling the
discharge of the unit, the hydro utility has no guarantee of
operating at its design water head. Therefore, it seems more
important to obtain an accurate model along the higher
efficiency path for the unit, according to different values of
head, rather than concentrating the efforts around the design
point.
Although differences between the 9-coefficient and the 6coefficient models were small, the best fitting was obtained
for the first model, with practically no differences on whether
imposing or not a maximum at the design point. An additional
analysis showed that the deviation in the generation values
whether considering an average unit efficiency or the exact
hill curve efficiency for the operating conditions were small
for centralized short term dispatch purposes. However, when
scheduling the operation for a specific plant with a given
weekly target from the system operator, gains may be
meaningful when considering an accurate model for the hill
curves.
Finally, the methodology proposed in this paper can be
used as a support tool for hourly or daily dispatch models that
consider the hydro units efficiency curves.
VII. REFERENCES
[1]

[2]
[3]
[4]

[5]

[6]
[7]

[8]

M.E.P. Maceira, L.A. Terry., F.S. Costa , et al. , Chain of optimization


models for setting the energy dispatch and spot price in the Brazilian
system, Proceedings of the Power System Computation Conference PSCC02, Sevilla, Spain, June 24-28, 2002.
O. B. Fosso, A. Gjeslvik, A. Haugstad, B. Mo, I. Wangensteen,
Generation scheduling in a deregulated system. The Norwegian case,
IEEE Trans. Power Systems, vol. 14, no. 1, Feb. 1999, pp. 75-81.
Z. K. Shawwash, T. K. Siu, S. O. D. Russel, The B.C. Hydro short term
hydro scheduling optimization model, IEEE Trans. Power Systems, vol.
15, no. 3, Aug. 2000, pp. 1125-1131.
A. L. Diniz, M. E. Maceira and L. A. Terry, Modelagem da funo de
produo energtica das usinas hidroeltricas nos modelos de otimizao
do planejamento a curto prazo e da programao do despacho de sistemas
hidrotrmicos interligados, Proceedings of the IX Symposium of
Specialists in Electric Operational and Expansion Planning SEPOPE,
Rio de Janeiro, Brazil, May 2004.
E. C. Finardi, E. L. da Silva and C. A. Sagastizbal, Solving the unit
commitment problem of hydropower plants via Lagrangian relaxation and
sequential quadratic programming, accepted for publication in
Computational and Applied Mathematics, 2004.
A. Arce, T. Ohishi and S. Soares, Optimal dispatch of generating units of
the Itaip hydroelectric plant, IEEE Trans. Power Systems, vol. 17, no. 1,
Feb. 2002, pp. 154-158.
F. Manzanedo, J. L. Castro and M. Perez-Donsion, Application of
evolutionary techniques to short-term optimization of hydrothermal
systems, Proceedings of the 2000 International Conference on Power
Systems Technology PowerCon 2000, 4-7 Dec. 2000, v.3, pp. 15391544.
L. A. M. Fortunato, T. A. A. Neto, J. C. R. Albuquerque and M. V. F.
Pereira, Introduo ao planejamento da expanso e operao de sistemas
de produo de energia eltrica, Niteri: Universidade Federal Fluminense,
EDUFF, 1990.

[9]
[10]
[11]
[12]
[13]

[14]

[15]
[16]

[17]
[18]

A. H. Frampton and G. D. Floyd, Factors in the economic supply of


energy in hydroelectric systems, AIEE Transactions, vol. 66, Jan. 1947,
pp. 1117-1125.
R. B. Allen and S. G. Bridgeman, Dynamic programming in hydropower
scheduling, J. Water Resources Planning and Management, vol. 112, no.
3, July 1986, pp. 339-351.
W. T. Alley, Hydroelectric plant capability curves, IEEE Trans. Power
Apparatus and Systems, v. 96, no. 3, May/June 1977, pp. 999-1003.
O. Nilsson and D. Sjelvgren, Variable splitting applied to modelling of
start-up costs in short term hydro generation scheduling, IEEE Trans.
Power Systems, vol. 12, no. 2, May 1997, pp. 770-775.
A.L. Diniz, M.E. P. Maceira, C.A. Sagastizabal, E.C. Finardi et al,
Hydro unit-commitment via lagrangian relaxation. Application to the
Brazilian optimization model for short term scheduling of hydrothermal
interconnected systems DESSEM, Proceedings of the VIII Symposium
of Specialists in Electric Operational and Expansion Planning SEPOPE,
Rio de Janeiro, Brazil, May 2004.
U. Navon, I. Zur and D. Weiner, Simulation model for optimising energy
allocation to hydro-electric and thermal plants in a mixed thermal / hydroelectric power system, IEE Proceedings, v. 135, part C, no. 3, May 1988,
pp. 182-188.
M. Breton, S. Hachem and A. Hammadia, Account for losses in the
optimization of production of hydroplants, IEEE Trans. Energy
Conversion, vol. 19, no. 2, June 2004, pp. 346-351.
E. I. B. Ang, J. R. Teixeira F., J. C. Negri, P. G. C. Maimone and S. N.
Barillari, Determinao do rendimento ponderado mdio operacional de
turbinas de usinas hidreltricas, Proceedings of the XVII Seminrio
Nacional de Produo e Transmisso de Energia Eltrica SNPTEE, Oct.
2003.
N. R. Draper and H. Smith, Applied Regression Analysis, New York, John
Wiley & Sons, 1998.
A. L. Diniz, T. N. Santos, M. E. P. Maceira, Short term security
constrained hydrothermal scheduling for large scale systems considering
transmission losses, Proceedings of the IEEE/PES Transm. Distr. Conf.
Expos. Latin America, June 2006.

Andre Luiz Diniz received the BSc. in Civil Engineering and MSc. in
Operations Research respectively in 1996 and 2000, at the Federal University of
Rio de Janeiro (UFRJ), and is working toward his DSc. degree in optimization at
COPPE / UFRJ.
Since 1998 he has been a researcher at CEPEL, the Brazilian Electrical
Power Research Center, where he has been working in mathematical models for
the mid term and short term hydrothermal scheduling, including hydro and
thermal unit commitment. He is also an assistant professor at the Institute of
Mathematics and Statistics at the State University of Rio de Janeiro, UERJ..
Pedro Paulo Ielo Esteves obtained the BSc in Civil Engineering in ITA,
So Carlos, in 1986, and is working towards his Msc degree at IMPA, Instituto
de Matemtica Pura e Aplicada. He has worked as a researcher at CEPEL,
Centro de Pesquisas de Energia Eltrica, from 2003 to 2005, where he
developed models for the short term operation planning, specially in the hydro
unit commitment problem.
Claudia Alejandra Sagastizbal.. has a doctoral degree in Applied
Mathematics and the Habilitation a diriger des recherches degree, obtained in
1993 and 1998 respectively, both at the University of Paris I, PantheonSorbonne, in France. She has worked at Inria, the French National Institute for
Research in Computer Science and Control, and also taught in various
universities and Grandes Ecoles' in Argentina, France and Brazil. She
collaborated on industrial projects with Electricit de France and Renault-France
and is currently working at CEPEL on energy optimization problems. She is a
world-leading specialist in nondifferentiable optimization, and has published 3
books and more than 40 papers in international journals.

You might also like