Optimal Operation Scheduling of Pumped Storage Hydro
Power Plant and Thermal Power Plants in Power System with
a Large Penetration of Photovoltaic Generations
Ryota Aiharaa and Akihiko Yokoyamab
a
Department of Electrical Engineering and Information Systems, The University of Tokyo, Japan
b
Department of Advanced Energy, The University of Tokyo, Japan
aihara@syl.t.u-tokyo.ac.jp
Abstract — In recent years, a substantial amount of
photovoltaic (PV) generations have been installed in power
systems. However, the power output from the PV is random
and intermittent in nature. Therefore, the PV generation
poses many challenges to the power system operation. To
evaluate the impact of the behavior of PV on power supply
reliability, we have developed a power supply reliability
evaluation model considering a large integration of PV
generation into power system. As a result, the power supply
reliability is getting worse as the PV penetration level
increases. To solve this issue, we have proposed that pumped
storage hydro power plant (PSHPP) is used effectively to
improve the reliability. In this paper a cooperative
scheduling method for the PSHPP and thermal power plants
that makes it possible to improve both reliability and
economy is presented.
Keywords — Monte Carlo Simulation, Optimal Scheduling,
Pareto Optimal Solutions, Photovoltaic Generation, Pumped
Storage Hydro Power Plant, Surplus Power Problem
I. INTRODUCTION
In recent years, a large amount of photovoltaic (PV)
and wind power generations have been installed in power
systems around the world. However, Japan is not an
appropriate site to install a large amount of wind power
generations. Therefore, the Japanese government aims at
installing a large amount of PV in power system.
Since the power output from the PV is random and
intermittent in nature, the PV generation poses many
challenges to the power system operation. When the load
demand is very small and at the same time the PV
generates a large amount of power, the balance between
power supply and demand cannot be maintained. This
problem is called “surplus power problem". In addition to
the surplus power problem, it is also necessary to
consider the intermittent and uncertain output
characteristic of PV.
Therefore, various kinds of countermeasures have been
considered when a substantial amount of PV generation is
installed into a power system. Though Battery Energy
Storage System (BESS) could be considered as a solution
of these issues, their high cost is preventing them from
being considered as a solution in most situations. This
paper proposes an effective use of Pumped Storage
Hydro Power Plant (PSHPP). PSHPP is installed
originally for hot reserve of the power system and load
leveling. The energy is stored in the form of water
pumped up from a lower reservoir to an upper reservoir.
The PSHPP is usually operated as a generator in the
daytime and a pump in the nighttime. A new operation
scheduling of PSHPP is proposed to solve the problem in
this paper. It will be operated with the proposed pattern
solving the surplus power problem caused by the PV
generation. The authors have already proposed how the
PSHPP operation pattern is changed effectively to
improve the power system reliability in case of a large
integration of PV into the power system [1]. However,
the scheduling method for obtaining the optimal PSHPP
operation pattern that makes it possible to improve both
reliability and economy has not been developed yet. In
this paper, we propose a new method for planning PSHPP
operation pattern considering the surplus power problem,
power supply reliability and fuel cost reduction, which is
represented by pareto optimal solutions. The power
supply reliability and the fuel cost are estimated for each
PSHPP operation plan by using Monte Carlo simulation.
II. SIMULATION MODEL
A. Pumped Storage Hydro Power Plant
In PSHPP, the energy is stored in upper reservoir in
the form of water. Therefore, it should be operated within
the capacity of the upper reservoir. In this paper, the
amount of the stored water in the upper reservoir is
considered as energy. It can be operated 6 to 8 hours
continuously when the operation starts from the full
capacity of the upper reservoir. Therefore, the capacity of
the upper reservoir is assumed to be 7-hour operation of
the rated power at generator mode.
In general, there are two types of PSHPP, fixed speed
one and adjustable speed one. The fixed speed PSHPP
cannot adjust the input power when it operates as a pump.
The adjustable speed one can adjust the input power. It is
coming into the limelight because of its adjustable
characteristics in the pump mode [2]. Therefore, the
PSHPP is assumed to be an adjustable speed PSHPP in
this paper.
 100
90
80

PV Output [%]
70
60
50
40
30
20
10
0
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
Time

Fig. 2 Theoretical PV Output

D. Monte Carlo Simulation

In this paper, we consider three types of probabilistic

uncertainty concerning load demand, power output of PV
Fig. 1 Flowchart of Power Supply Reliability and generator failure.
Calculation The load demand model with a randomly fluctuating
component is derived from (1).
B. Reliability Evaluation Model
L dt  Lgt  F (1)
The proposed evaluation method of the power supply
reliability is summarized as a flowchart in Fig.1. In this where Ldt is load demand at time t considering
paper, it is assumed that the operation schedule of PSHPP probabilistic fluctuation. Lgt is load demand at time t
is planned for a week. The simulation period is a week given by the power system model and F is the random
and its sampling time is 1 hour. First, based on the number based on the normal distribution where the
available data, the generation schedule for 7 days is average is 1 and the standard deviation is 3%.
determined. The available data consist of load demand
given by the power system model, the theoretical PV The power output of PV generation is also modeled
generation output and PSHPP operation pattern. The with a randomly fluctuating component.
generation schedule is determined by solving
optimization problem, which minimizes the fuel cost by PVdt  PVgt  F (2)
using Dynamic Programming (DP). Next, Monte Carlo
simulation is carried out considering uncertainties of load
demand, PV output and generator failure. During the where PVdt is PV output at time t considering
period of the simulation, when the power supply and probabilistic fluctuation, PVgt is PV output at time t given
demand imbalance occurs, the thermal generator outputs by the theoretical output and F is the random number
are adjusted and PSHPP is operated to avoid the based on the normal distribution considering weather
imbalance. If it cannot be avoided, this period is regarded changes shown in Table 1.
as supply interruption. We assume that both supply
shortage and surplus power supply are defined as the Table 1 Random Characteristic of Weather Changes
supply interruption. Weather Average Standard Deviation [%]
Sunny 1 3
C. Generation Scheduling Cloudy 0.5 20
Rainy 0.1 0.03
In this paper, it is assumed that nuclear and hydro
power plants are always operated at the rated power
The weather change is simply modeled at the
output. As a result, only the fuel cost of thermal plants is
probability of 1/3 at time t. It is assumed that the weather
considered. The feed-in tariff is not included in the fuel
is independent between before and after time t. The
cost. The merit order method is used in the unit
theoretical PV outputs for three kinds of weathers are
commitment. The merit order is determined by using the
shown in Fig. 3.
fuel cost of generator at the rated output. For
simplification, the shutdown cost is not considered here.
The PV output is assumed to be theoretical in generation
scheduling because power system operator cannot predict
its output precisely at this time. The theoretical power
output of PV is shown in Fig. 2.
 100
90
Sunny load demand is constant, the weekly fuel cost is not
80 Cloudy
changed from the cost which determined by DP. However,
PV Output [%]

70 Rainy
60 the PV output and the load demand could fluctuate
50
40 randomly in Monte Carlo simulation. Therefore, the load
30
20
dispatch of generators is changed and its operational cost
10 rises or falls at each trial. In this paper, the weekly fuel
0
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 cost is defined as the mean value of fuel costs of all trials
Time

Fig. 3 PV Output for Three Kinds of Weather excluding the trial where the generator failure occurs.
This is because the fuel cost increase is the unexpected
The generator failure occurs during Monte Carlo factor due to the generator failure. In other words, the
simulation at the failure rate λ described by (3). In (3), outage cost is not determined uniquely.
MTBFi is the mean time between failure of generator i
and MTTRi is the mean time to repair of generator i
III. OPTIMAL OPERATION SCHEDULING
1 1 (3)
i  , i 
MTBFi MTTRi A. Pareto Optimal Solutions

During Monte Carlo simulation, if the generator failure
occurs at time t, the other generators are re-dispatched
immediately. Figure 4 shows how the generators are re-
dispatched when generator G3 is tripped at time t, for
example. First, Generator G1, G2 and G4 are re-
dispatched immediately. If imbalance of power supply
and demand cannot be avoided by the re-dispatching,
PSHPP is operated immediately to avoid it as an
emergency control. In the next time t+1, the remaining
generators are committed to compensate for the
unavailable generator. The tripped generator is restored at
the restoration rate μ after time t+1.
Generator Failure
First Step
Supply Demand
Shortage
Supply
In general, improvement of power system reliability
and reduction of operational cost are competing.
Optimization of the PSHPP scheduling has no unique
solution. This type of problem is a multi-objective
optimization problem. In this paper, pareto optimal
solutions are obtained.
B. Quantization of PSHPP operation pattern
The operation scheduling of PSHPP and thermal
power plants are quantized as shown in Fig. 5. PSHPP
operation pattern is quantized by 6 steps at each time t.
Generators Re-dispatch
Thermal power plants operation is represented as a
Second Step number of surplus generators from the cost minimize case
Supply Demand
determined by dynamic programming [3].
Emergency

Control

PSHPP
G4

G5

0 25% 50% 75% 100%

Power Output of PSHPP

Mode Stop Generator Mode

G4

G4

・・・
G3

Load Demand
Load Demand

Load Demand
Load Demand

G4
Maximum Output

Maximum Output
G2

168hours
G2

1 2 3 4 5 6 7 8 9 10
G2

G2

・・・
Pump

Time
-100%
G1

G1
G1

G1

PSHPP 5 4 4 3 1 0 0 1 1 5 ・・・ 4

Thermal 0 2 2 0 0 1 0 1 1 0 ・・・ 0
Time t Time t+1

Fig. 4 Generator Re-dispatching Power Output of PSHPP

Surplus Generators

The power supply reliability is evaluated through Fig. 5 Quantization of Operation Scheduling
LOLP (Loss of Load Probability) index which can be
computed from (4). LOSS is the total supply interruption C. Optimization Problem
period. N is the number of trial. T is the simulation period.
The trial is repeated 10,000 times until LOLP is The optimization problem for operation scheduling of
converged to the steady-state value. PSHPP and thermal power plants is then formulated as
follows:
LOSS (4)
LOLP 
N T Control variable:
 Ps Ps2 Ps3 ... Pst 
The weekly fuel cost is calculated at each trial during v 1
Ts3 ... Tst 
(5)
Monte Carlo simulation. If the power output of PV and Ts1 Ts2
Objective functions is not good at local search. Therefore, this paper proposes
Minimize: that GA and TS are combined to obtain pareto optimal
N T I PSHPP operation patterns. Tabu search is a local search
f1   technique which enhances the performance using
n 1 t 1 i 1 memory structures named “Tabu List”. Once a potential
{u F i ( Pt i,n )  uti,n (1  u(it 1),n ) S i }
i
t ,n (6) solution has been determined, it is marked as "taboo" so
that the algorithm does not visit that possibility
repeatedly.
N An algorithm overview of the proposed method is
 Lo n shown in Fig. 6.
f2  n 1
(7)
N T

Subject to:

Upper reservoir of the PSHPP

0  Ust  U max (8)
0  Umt ,n  U max (9)

Power output of the PSHPP

 Pumax  Pst  Pumax (10)
 Pumax  Put ,n  Pumax (11)

Power output of the thermal plants

 Pmin
i
 Pt i,n  Pmax
i
(12)

Fixed power plants

PN  constant (13)
PH  constant (14)

where Pst is the PSHPP output at time t. Put,n is the

PSHPP output at time t, trial n. Tst is the no. of surplus
thermal generators at time t. N is the no. of trials. T is the
simulation time. I is the number of thermal power plants.
Pit,n is power output of generator i at time t, trial n. uit,n is
state variable of generator i at time t, trial n (1: committed
0: stopped). Si is startup cost of generator i. Lon is the
supply interruption period at trial n. Ust is the capacity of
upper reservoir of the PSHPP at time t in schedule. Umt,n
is the capacity of upper reservoir of the PSHPP at time t,
trial n in Monte Carlo simulation. Umax is the total
capacity of upper reservoir of the PSHPP. Pimin is
minimum output of generator i. Pimax is maximum output
of generator i. PN is power output of nuclear power plant.
PH is power output of hydro power plant.

In this study, operation schedule of PSHPP and

thermal plant is determined by Genetic Algorithm (GA)
and Tabu Search (TS). GA is one of the evolutionary
algorithms, which generates solutions to an optimization
problem using techniques inspired by natural evolution,
such as inheritance, mutation, selection and crossover. It
generates multiple solutions simultaneously at each Fig. 6 Algorithm Overview
generation, and it is good at global search. However GA
 IV. SIMULATION CONDITION In both summer and spring seasons, it can be seen that
the pareto optimal solutions are located in a narrow range
A. Simulation Model of Cost-LOLP space when PV penetration is 0MW. The
The simulation is carried out on the modified IEEE 24- pareto optimal solutions do not include various kinds of
bus Reliability Test System (RTS) [4]. The system is operations of PSHPP because there is no PV output. On
modified by adding a 300MW PSHPP in addition to the the contrary, the pareto optimal solutions includes various
existing hydro power plants. Since this paper focuses on kinds of operation patterns when PV penetration is
the power supply reliability evaluation, failures in 1000MW in both summer and spring seasons.
transmission network are not considered. B. Operation Schedule of PSHPP
B. Simulation Condition The obtained operation schedule of PSHPP is shown in
Fig. 9. The best solutions concerning LOLP and the fuel
cost are discussed here.
3000

2500 Best LOLP Best Cost

Power Demand [MW]

Reservoir [MWh]
2000

Cap. of Upper
2000
1500
1500 1000
500
1000
0
500
Sa t Sun Mon Tue Wed Thu Fr i
Summer (August) Day
Spring (May)
0
Sa t Sun Mon Tu e We d Thu Fr i
(a) PV 0MW in Summer
Day

Fig. 7 Load Demand Curve Best LOLP Best Cost

Reservoir [MWh]

2000
Cap. of Upper

1500
Figure 7 depicts two different load conditions for two 1000
seasons, i.e. summer (August) and spring (May). The 500
0
system peak load is 3200 MW. In the simulation, the Sa t Sun Mon Tue Wed Thu Fr i
Day
amount of PV penetration is set to be 0MW and 1000MW.
(b) PV 1000MW in Summer

Best LOLP Best Cost

Reservoir [MWh]

2000
Cap. of Upper

V. SIMULATION RESULTS 1500

1000
A. Pareto Optimal Solutions 500
0
The pareto optimal solutions obtained by the proposed Sa t Sun Mon Tue Wed Thu Fr i
Day
optimization method are shown in Fig. 8. (c) PV 0MW in Spring
3.9
PV0MW
Best LOLP Best Cost
Reservoir [MWh]

3.8 2000
Cap. of Upper
Weekly Fuel Cost [Million $]

PV1000MW
3.7 1500
3.6 1000
3.5 500
3.4 0
Sa t Sun Mon Tue Wed Thu Fr i
3.3 Day
3.2
(d) PV 1000MW in Spring
3.1
3 Fig. 9 PSHPP operation patterns
2.9
0 0.05 0.1 0.15 0.2 0.25
LOLP
(a) Summer In summer, the PSHPP is operated as a pump in the
2
PV0MW
nighttime and as a generator in the daytime on weekday
1.9
to supply the heavy summer load when PV penetration is
Weekly Fuel Cost [Million $]

PV1000MW
1.8
1.7 0MW. The role of the PSHPP in this case is load leveling
1.6 within a week. It is the conventional role of PSHPP in the
1.5
power system. Compared with the operation patterns
1.4
1.3 between LOLP best case and cost best case, they are not
1.2 so much difference. The near optimal operation pattern is
1.1 determined uniquely in summer season when PV
1
0 0.05 0.1 0.15 0.2 0.25 penetration is 0MW. In contrast, when the PV penetration
LOLP
increases to 1000 MW in summer, the PSHPP is operated
(b) Spring as a pump in the daytime and as a generator in the
Fig. 8 Pareto Optimal Solutions nighttime in the LOLP best case in spite of no surplus
power. However, PSHPP is hardly operated in the cost
best case.
In spring, the PSHPP is hardly operated when the PV
penetration is 0MW because the operation of the PSHPP
does not contribute much to the fuel cost reduction. When
the PV penetration increases to 1000 MW, PSHPP is
hardly operated in the cost best case, either. The PSHPP
has to be operated as a pump in the daytime both on the
weekday and weekend in the LOLP best case. This
pattern is similar to that of summer pattern.
C. Generation Scheduling
The weekly thermal generation schedules for the
obtained PSHPP operation patterns are shown in Fig. 10.
The LOLP best case in both summer and spring season
when PV penetration is 1000MW are discussed here. In
Fig. 10, the PSHPP pushes up the load demand in the
daytime and pushes it down in the nighttime. The
deducted load demand shown by solid black line falls
below the minimum output of thermal plants in the
daytime, which shows that the surplus power occurs.
However, in this case, the PSHPP can be operated for
both PV output decrease and increase by emergency
operation. It is concluded that modification of the
operation pattern of the PSHPP is effective for the
surplus power problem.
Maximum Output of Thermal
Hydro
Load Demand (w/o PV, w/o Pump)
Load Demand (with PV, w/o Pump)
Power Output / Load Demand [MW]
VI. CONCLUSION
This paper presents a cooperative scheduling method
that optimizes both PSHPP operation and hot reserve
capacity of thermal power plants. The method makes it
possible to improve both reliability and economy in the
power systems with a large integration of PV. The
simulation results show that the proposed method is very
effective in terms of the reliability of power system
operation. The total fuel cost of thermal power plants
increases in order to secure the power supply reliability
using operation of the PSHPP in power system with a
large penetration of PV.
REFERENCES
[1] R. Aihara, A. Yokoyama, F. Nomiyama and N. Kosugi,
“Impact of Operational Scheduling of Pumped Storage
Power Plant Considering Excess Energy and Reduction of
Fuel Cost on Power Supply Reliability in a Power System
with a Large Penetration of Photovoltaic Generations”, in
Proc. 2010 International Conference on Power System
Technology, Hangzhou, China, Oct. 2010.
[2] T. Kuwabara, A. Shibuya, H. Furuta, E. Kita and K.
Mitsuhashi, ”Design and dynamic response characteristics
of 400 MW adjustable speed pumped storage unit for
Ohkawachi Power Station, ” IEEE Transactions on
Minimum Output of Thermal
Energy Conversion, vol. 11, No. 2, 1996.
Nuclear
Load Demand (with PV, with Pump)
[3] Allen J. Wood and Bruce F. Wollenberg, "Power
Generation, Operation, and Control", John Wiley& Son,

4000
1983
3500
3000
[4] Reliability Test System Task Force of the Application of
2500
Probability Methods subcommittee, ”IEEE
2000
RELIABILITY TEST SYSTEM,” IEEE Transactions on
1500 Power Apparatus and Systems, vol. PAS-98, No. 6, 1979.
1000
500
0
Sa t Sun Mo n Tu e We d Th u Fr i
Day
(a) PV 1000MW in Summer, for Best LOLP

Maximum Output of Thermal Minimum Output of Thermal

Hydro Nuclear
Load Demand (w/o PV, w/o Pump) Load Demand (with PV, with Pump)
Load Demand (with PV, w/o Pump)
Power Output / Load Demand [MW]

4000
3500
3000
2500
2000
1500
1000
500
0
Sa t Sun Mo n Tu e We d Th u Fr i
Day
(b) PV 1000MW in Spring, for Best LOLP
Fig. 10 Weekly Generation Schedule