Decision support model for weekly operation of

hydroelectric reservoirs by stochastic nonlinear

optimization

A. Ramos, S. Cerisola, J.M. Latorre

Universidad Pontificia Comillas
Contents

¾ Introduction
¾ System modeling
¾ Model description
¾ Case study
¾ Conclusions

11th Conference on Stochastic Programming (SPXI) - 2

Renaissance of hydro scheduling models

• Nowadays, under a deregulated framework electric

companies manage their own generation resources
and need detailed operation planning tools
• In the next future, high penetration of intermittent
generation is going to force the electric system
operation
• Hydro and storage hydro plants are going to play a
much more important role due to their flexibility and
complementary use with intermittent generation

11th Conference on Stochastic Programming (SPXI) - 3

Medium term model (i)

• Hydroelectric model deals only with hydro plants

• Hydrothermal model manages simultaneously both
hydro and thermal plants
• Thermal units considered individually. So rich marginal
cost information for guiding hydro scheduling
• No aggregation or disaggregation process for hydro
input and output is needed
• It is very difficult to obtain meaningful results for each
hydro plant because it requires a huge amount of data
and the complexity of hydro subsystems

11th Conference on Stochastic Programming (SPXI) - 4

Medium term model (ii)

• Determines:
– the optimal yearly operation of all the thermal and hydro power
plants
– taking into account multiple basins and multiple cascaded
reservoirs connected among them
• Cost minimization model because the main goal is
medium term hydro operation

11th Conference on Stochastic Programming (SPXI) - 5

Contents

¾ Introduction
¾ System modeling
¾ Model description
¾ Case study
¾ Conclusions

11th Conference on Stochastic Programming (SPXI) - 6

Demand

• Weekly demand with two load levels (peak and off-peak

each week)

40000
Demand [MW ]

30000

20000
10000

0
1 13 25 37 49

11th Conference on Stochastic Programming (SPXI) - 7

Hydro subsystem

• Different modeling approach for hydro reservoirs

depending on:
– Owner company
– Relevance of the reservoir
• Reservoirs belonging to other companies modeled in
energy units [GWh]
• Own reservoirs modeled in water units [hm3, m3/s]
• Important reservoirs modeled with water head effects
• Very diverse system:
– Hydro reservoir volumes from 0.15 to 2433 hm3
– Hydro plant capacity from 1.5 to 934 MW

11th Conference on Stochastic Programming (SPXI) - 8

Stochasticity sources

• Natural hydro inflows (clearly the most important factor

in Spanish electric system)
Year Hydro energy Index Probability of being
Available [TWh] exceeded [%]
2001 32.9 1.13 32
2002 20.9 0.72 87
2003 33.2 1.15 30
2004 22.7 0.79 80
2005 12.9 0.45 100
2006 24.0 0.83 70

• Changes in reservoir volumes are significant because

of:
– stochasticity in hydro inflows
– chronological pattern of inflows and
– capacity of the reservoir with respect to the inflows

11th Conference on Stochastic Programming (SPXI) - 9

Historical natural inflows
1400

1975
1200
1976
1977
1978
1000 1979
1980
1981
1982
Natural Inflows [m3/s]

800
1983
1984
1985
600 1986
1987
1988
1989
400
1990
1991
1992
200 1993
1994
1995
1996
0
1997
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51

-200
Week

11th Conference on Stochastic Programming (SPXI) - 10

Scenario tree generation

• A multivariate scenario tree is obtained by neural gas

clustering technique that simultaneously takes into
account the main stochastic series and their spatial and
temporal dependencies.
• Very extreme scenarios can be artificially introduced
with a very low probability
• Number of scenarios generated enough for yearly
operation planning

11th Conference on Stochastic Programming (SPXI) - 11

Natural inflows: scenario tree

1400

1200

Serie1
1000
Serie2
Serie3
Serie4
Natural Inflows [m3/s]

800 Serie5
Serie6
Serie7
Serie8
600
Serie9
Serie10
Serie11
400 Serie12
Serie13
Serie14
Serie15
200
Serie16

0
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51

-200
Week

11th Conference on Stochastic Programming (SPXI) - 12

Contents

¾ Introduction
¾ System modeling
¾ Model description
¾ Case study
¾ Conclusions

11th Conference on Stochastic Programming (SPXI) - 13

Constraints: Generation and load balance

Generation of thermal units

+ Generation of hydro units
– Consumption of storage hydro units
= Demand

11th Conference on Stochastic Programming (SPXI) - 14

Constraints: Minimum and maximum
operating hours of thermal units
• Introduced to model:
– Unavailability of thermal units
– Domestic coal subsidies
– CO2 Emission allowances
– Capacity payments
• They are not separable by period

minimum ≤ Yearly operation hours of each thermal unit for

each scenario ≤ maximum

minimum ≤ Average yearly operation hours of each

thermal unit ≤ maximum

11th Conference on Stochastic Programming (SPXI) - 15

Constraints: Water balance

Reservoir volume at the beginning of the period

+ Natural inflows
– Spills from the own reservoir
+ Spills from upstream reservoirs
+ Turbined water from upstream hydro plants
+ Pumped water from downstream hydro storage plants
– Turbined and pumped water from the own reservoir
= Reservoir volume at the end of the period

11th Conference on Stochastic Programming (SPXI) - 16

Constraint: Water head effects

• Power generation is the product (nonlinear function) of

the flow and the production function
P = Q x PF

• Production function PF depends linearly on water head

P ro d u c tio n fu n c tio n [M W /m 3 /s ] 0.300

0.250

PF = α Hpp 0.200

0.150

0.100

0.050

0.000
0 20 40 60 80 100 120
Water head [m]

11th Conference on Stochastic Programming (SPXI) - 17

Constraint: Volume as a function of the head

• Reservoir volume depends quadratically (nonlinearly)

on water head
V = β + β’ Hrr+ β” Hrr22

3000

2500
Volume [hm3]

2000

1500

1000

500

0
0 20 40 60 80 100 120
Water head [m]

11th Conference on Stochastic Programming (SPXI) - 18

Constraint: Water heads

Water head of the reservoir = forebay level – reference level

Water head of the plant = forebay level of the reservoir –

tailrace level of the plant

Tailrace level of the plant = max [forebay level of downstream

reservoir, reference tailrace level of the plant]

11th Conference on Stochastic Programming (SPXI) - 19

Constraint: operation limits

Reservoir volumes between limits for each hydro reservoir

Power operation between limits for each unit

11th Conference on Stochastic Programming (SPXI) - 20

Multiobjective function

• Thermal plant variable costs

• Penalties introduced in the objective function for

softening several additional constraints:
– Final reservoir volumes
– Exceeding operating rule curves (minimum and maximum)
– Minimum and maximum yearly operation hours of thermal units

11th Conference on Stochastic Programming (SPXI) - 21

Type of optimization problem

• Deterministic approaches:
– Network Flows
– LP
– NLP
– MILP
• commitment of thermal or hydro units
• piecewise linear approximation of water head effects
• Stochastic approaches:
– Stochastic Dynamic Programming (SDP)
– Stochastic Linear Programming. Decomposition approaches
(Benders, Lagrangean Relaxation, Stochastic Dual Dynamic
Programming)
– Stochastic Nonlinear Programming

11th Conference on Stochastic Programming (SPXI) - 22

Solution algorithm

• Algorithm:
– Successive LP
– Direct solution by a NLP solver

• Very careful implementation

– Scaling of variables
– Use of simpler expressions
– Initial values and bounds for all the nonlinear variables
computed from the solution provided by linear solver CPLEX
10.2 IPM
– Nonlinear solver CONOPT3

11th Conference on Stochastic Programming (SPXI) - 23

Model implementation

• General hydro topology

• Spreadsheet-based graphical interface
• GAMS-based optimization model

11th Conference on Stochastic Programming (SPXI) - 24

Contents

¾ Introduction
¾ System modeling
¾ Model description
¾ Case study
¾ Conclusions

11th Conference on Stochastic Programming (SPXI) - 25

Case study

• Spanish electric system

– 130 thermal units
– 3 main basins with 50 hydro reservoirs/plants and 2 pumped
storage hydro plants
– 16 scenarios

• Problem size:
– 271887 constraints
– 442239 variables
– 1611184 non zero
elements
– 12832 nonlinear
variables
– 8020 nonlinear
constraints

11th Conference on Stochastic Programming (SPXI) - 26

Hydro plant operation

• Relative error in the energy generated for each hydro

plant between LP and NLP approaches

8.0%
6.0%
4.0%
Relative error [%]

2.0%
0.0%
-2.0% 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49

-4.0%
-6.0%
-8.0%
-10.0%
Hydro plant

11th Conference on Stochastic Programming (SPXI) - 27

 Hydro reservoir operation (i)
RVA Mín RVA Máx + Probable C. Garantía C. Resguardo Seco 5 %
Seco 25 % Medio Húmedo 25 % Húmedo 95 %

3000.0

2500.0

LP 2000.0

Reservoir Volume [hm ]

3
solution
1500.0

1000.0

500.0

0.0
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51
Week

RVA Mín RVA Máx + Probable C. Garantía C. Resguardo Seco 5 %

Seco 25 % Medio Húmedo 25 % Húmedo 95 %

3000.0

2500.0

2000.0
Reservoir Volume [hm ]
3

1500.0
NLP
1000.0
solution
500.0

0.0
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51
Week 11th Conference on Stochastic Programming (SPXI) - 28
 Hydro reservoir operation (ii)
RVA Mín RVA Máx + Probable C. Garantía C. Resguardo Seco 5 %
Seco 25 % Medio Húmedo 25 % Húmedo 95 %

1400.0

1200.0

LP 1000.0

Reservoir Volume [hm ]

solution

3
800.0

600.0

400.0

200.0

0.0
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51
RVA Mín RVA Máx + Probable C. Garantía C. Resguardo Seco 5 % Week
Seco 25 % Medio Húmedo 25 % Húmedo 95 %

1400.0

1200.0

1000.0
Reservoir Volume [hm ]
3

800.0

600.0
NLP
400.0
solution
200.0

0.0
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51
Week 11th Conference on Stochastic Programming (SPXI) - 29
Contents

¾ Introduction
¾ System modeling
¾ Model description
¾ Case study
¾ Conclusions

11th Conference on Stochastic Programming (SPXI) - 30

Summary

• Medium term hydrothermal model

• Nonlinear water head effects modeled for relevant

reservoirs

• Stochastic nonlinear optimization problem solved

directed by a nonlinear solver given a close initial
solution provided by a linear solver

11th Conference on Stochastic Programming (SPXI) - 31

Decision support model for weekly operation of
hydroelectric reservoirs by stochastic nonlinear
optimization

A. Ramos, S. Cerisola, J.M. Latorre

Universidad Pontificia Comillas

11th Conference on Stochastic Programming (SPXI) - 32