Reliability Assessment of a Power System with High

Penetration of Wind and Cascade Hydropower Plant

Considering Wind Correlation
Qian Dai, Pingliang Zeng, Qinyong Zhou, Baiqing Li Feng Zhao
China Electric Power Research Institute State Grid Jibei Electric Power Research Institute
Beijing, China Beijing, China
qdaihust@gmail.com zhaofeng730@126.com

Abstract—This paper presents a method to assess the impact on Therefore, when combined with the renewable generation,
system reliability of wind integration, considering coordinated they could store water during high wind periods, and increase
operation with cascade hydropower. The seasonal output when wind power goes down. Thereby, this
characteristics of river flows and correlation amongst different combination can improve the system ability to accommodate
wind farms are considered. Time sequential Monte Carlo renewable generation. Thermal power has stable output, which
simulation and Latin Hypercube Sampling techniques are used can satisfy the majority of load. In order to mitigate the
to preserve the characteristics of the time series of the wind operating risks and to increase the utilization of wind power, it
speed, water inflows, load, etc. Reliability indices are developed is important to coordinate the operation of different type of
and used to assess, the impact of wind energy production on
generating units in response to supply and demand variations.
system reliability. The proposed method is applied to northwest
grid of China. Results show the validity of the method.
It also becomes important to assess power system reliability
with multi-wind and cascaded hydro coordination.
Index Terms--Cascade hydropower plant, Latin Hypercube Several works has been conducted on the integration of
Sampling, power system reliability, time sequential Monte Carlo wind power and its impact on system reliability [1]-[6]. It is
simulation, wind correlation. important to accurately model wind output in studying the
impact of wind power integration on power systems. The
I. INTRODUCTION impact of wind speed correlation [2]-[4] and wake effect [5]-
Over the last decade, renewable generation, especially [6] on power system reliability has been studied. There has
wind generation, has seen rapid development and this trend is also been work reported on coordinated operation of wind
expected to continue for the foreseeable future. Wind farm and hydro thermal power station [7]-[12]. However, the
generation in China reached 87GW in total at the end of 2013 hydro power models used in such works is usually small
and is projected to reach 200GW by 2020. The majority of hydropower plants, such as run-of-river plants, and hydro
renewable generation is located in the northwest, north and plant with accumulation reservoir and cascade arrangement is
northeast areas of China. On the other hand, wind power not fully considered.
output is inherently variable because of its intermittent and In this paper, a time sequential Monte Carlo simulation
fluctuating characteristics. This is posing an increasing (TMCS) technique incorporating the coordination between
challenge to the reliable operation of power system as its hydro and wind energy sources has been developed for the
penetration increases. The amount of wind energy that can be evaluation of system adequacy. The North West Grid of China
absorbed by an electric power system can be greatly limited if is used as the test system consisting of large cascade
the available conventional units are not able to respond hydropower plants (CHPP) with high penetration of wind
quickly to wind power fluctuations. In some circumstances, a generation. The impact of the coordinated wind, hydro and
large amount of wind power has to be curtailed in order to thermal operation on the overall system adequacy is evaluated.
maintain power system reliability and security.
The remainder of this paper is organized as follows: wind
In order to solve wind integration problem, a mix of farm models are presented in Section II and cascade hydro
different generation technologies, such as hydro, thermal and ones are described in Section III, Section IV presents the
wind, with different operating characteristics is desired. Hydro sequential Monte Carlo simulation technique, followed by
power stations with a reservoir have an ability to change their case study in Section V. Finally, conclusions are presented in
power output quickly and act as an energy storage facility in Section VI.
the form of stored water as and when the system requires.
This work is supported by SGCC Special Project (CEPRI: XT71-12-
028), and the National High Technology Research and Development
Program of China (863 Program) (project No. 2011AA05A103).

978-1-4673-8040-9/15/$31.00 ©2015 IEEE

 II. WIND FARM MODEL which can be found in [6].
A. Wind speed model D. Wake Effect
Many studies have reported statistical tests on wind speeds Due to the limited area of wind farm, there is a relatively
using different distributions, such as Weibull, Rayleigh, etc [1]. close spacing between WTs. The output power of downstream
The Weibull distribution adequately represents the wind speed WT may be affected by the wake effect created by upstream
probability distribution, which is given by: WTs and result in power loss [5-6]. Therefore, it is essential to
consider the wake effect and to examine its impact on the
ki −1 overall wind power system reliability. The wake effect is
k ⎛X ⎞
f X i ( λi , ki ) = i ⎜ i ⎟
− ( X i λi )
ki

e , i = 1,...., N (1) considered in this paper with the wake effect, coefficient
λi ⎝ λi ⎠ assumed to be between 0.8 and 0.9. The sum of all WTs’
output power in a wind farm is denoted as wind farm’s power.
Where λi is the scale factor and ki is the shape factor of the ith
wind farm. The cumulative probability function is as: III. CASCADED HYDRO AND THERMAL MODEL
A. Cascade Hydro Generation Availability Model
FX i ( λi , ki ) = 1 − e
−( X i λi )
ki
The water inflow is very different with different weather
, i = 1,...., N (2)
condition. The probability of occurrence of each weather type
is assumed same [7]-[12]. The water inflow is simulated. The
B. Wind Speed Correlation ith CHPP water volume Vit is calculated at the end of the tth
hour. Rit and Qit are the water inflow and discharge rate.
The problem arises with the simulation of the wind speeds
in several wind farms when they are simultaneously
( )Δt
N
Vi t = Vi t −1 + ( Rit − Qit − S tp ,i ) Δt +∑ Qn (4)
t −τ n ,i t −τ n ,i
calculated, because a correlation between them may exist [2]- + Sn
n =1
[4]. The correlation wind speed of multi-wind farms can be
obtained by generating sets of wind speeds with given Weibull
distributions and correlation matrices, and can be summarized ⎧⎪Vi t − Vi max ,Vi t > Vi max
as follows:
S tp ,i = ⎨ (5)
⎪⎩0 ,Vi t ≤ Vi max
• Generate wind speeds by means of the Monte Carlo
simulation, according to the Weibull distribution of Vi min ≤ Vi t ≤ Vi max (6)
each wind farm.
Where Sp,i is the volume of the water spilled during tth hour, N
• The mean value of the wind speed of each wind farm is the set of upstream CHPP; Qnt-
τn,i
and Snt-
τn,i
are the
is subtracted from the wind speed obtained, and the th
discharge and spilled water flow from n upstream CHPP to
result of this operation is divided by the standard
the ith downstream CHPP after time delay τn,i, respectively.
deviation. With this operation, sets of uncorrelated
standardized and independent wind speed variables Vimin and Vimax are the lower limit and upper limit of ith CHPP
z=(z1, z2, . . . , zn)T are obtained. water volume. The discharge rate should satisfy the following
constraint:
• The correlated wind speed variables can be obtained
according to y = Lz + μ , where L is a lower Qimin ≤ Qit ≤ Qimax (7)
triangular matrix, μ is the mean of the new vector y= Where Qimin and Qimax are the lower and upper limit of ith
(y1, y2, . . . , yn)T with covariance matrix Ω y. The CHPP water discharge rate, respectively. And the minimum
matrix L is not unique, which is obtained by output power of the CHPP can be denoted as Pimin.
Cholesqui's technique.
Under this condition, if the wind-hydro-thermal generation
C. Wind Turbine Output Characteristic is higher than the system load, e.g. during high wind output
The output power of the wind turbine (WT) is primarily water can be stored in the reservoir. If total power output from
affected by the wind speed. A nonlinear relationship exists is in deficit, e.g. during low wind output, an appropriate extra
between the WT output power and the wind speed. The WT water flow can be released to meet the load. This operation
power output can be calculated [4]-[6] as: mode is denoted as coordination mode. The power output
from a hydro unit can be obtained as follows
⎧0 v ≤ vci or v ≥ vco Pi t = ηi Qit 3600 (8)
⎪
Pwt = ⎨ Pr ( B0 + B1v + B2 v 2 ) vci ≤ v ≤ vr (3)
Where ηi is the overall efficiency of the ith CHPP, and which
⎪ has considered the effect of the fall height.
⎩ Pr vr ≤ v ≤ vco
Where Pr is the rated power output of the WT, vci is cut-in B. Generator Model
wind speed; vr is rated wind speed; vco is cut-out wind speed. The thermal generator is modeled by a two states model
The constants B0, B1, B2 are determined by the vci, vr, and vco, [9], the available and unavailable time can be expressed as:
 1 In this paper, we focus on the reliability of generation system,
τ 1 = − ln γ 1 = −TMTTF ln γ 1 (9) i.e., the amount of load shedding is only related to the
λ available generation capacity and the amount of electric loads:
1 T ⎛ J ⎞
τ2 = − ln γ 2 = −TMTTR ln γ 2
W I
(10)
μ EENS = ∑ ⎜ ∑ Pjt + ∑ Pwt + ∑ Pi t − Lt ⎟Δt (14)
t =1 ⎝ j =1 w =1 i =1 ⎠
Where γ1 and γ2 are uniformly distributed random number over
the interval (0, 1). The availability of all system components Where T is the total simulation hours, which can be 8736 or
8760 hours based on the historical base data. J and W are the
are obtained for the simulated hour. λ is failure rate and μ is
number of thermal generation units and wind farms,
the repair rate. TMTTF and TMTTR are the mean time to failure (h)
respectively. Pjt and Pwt are the power output of jth thermal
and mean time to repair (h), respectively.
unit and the power output of wth wind farm, respectively. I is
C. Load Model the number of CHPP. Pit is the power output of ith CHPP.
The hourly simulated load Lt at time t is obtained from the C. Simulation of Wind and Hydro Power Coordination
mean load Lmt and its standard deviation σl, which can be
The overall TMCS procedure for generation adequacy
calculated by:
assessment is briefly described in the following steps.
Ltm = Ltw Ltd Lth PLmax (11) 1) Generate state duration series: The inverse transform
method and the distribution functions of the component
Lt = Ltm +N ( 0，σ l2 ) (12) failure and repair rates by equation (9) and (10) in a year for
each component is used (all generating units including WT,
Where Lwt is the weekly peak loads in percentage of the CHPP, and thermal stations). The initial state of each
annual peak, Ldt is the daily peak load in percentage of the component is assumed to follow uniform distribution.
weekly peak, Lht is the hourly load in percentage of the daily 2) Determine hourly states duration of generations:
peak, PLmax is the peak load of the whole year.
According to state residence time series, the hourly state of
IV. SEQUENTIAL MONTECARLO SIMULATION unit can be determined.
3) Generate Supply and Load Data: A chronological
A. Latin Hypercube Sampling hourly system load model is constructed and incorporated in
The CHPPs and wind farms are a component with time- the analysis. The Weibull distribution time series model is
varying parameters, such as reservoir level, water inflow, utilized to simulate hourly wind speeds. The chronological
volume of the water utilized to generate electricity and wind wind speed model is then constructed for the given time span.
speed. Therefore, time-dependent algorithms are required in Power outputs of WECS are calculated for each hour based
order to assess system reliability with highly time-varying on the hourly simulated wind speeds and the available WTG
components integrated. units during that hour. The power outputs of CHPP are
The TMCS is thus adopted in this paper. It is recognized as obtained by the water inflow, the unit status and the discharge
an effective algorithm to analyze problems of time dependent flow which can be found in Section III.A.
reliability. The sampling data is very important to the 4) Calculate EENS for Coordination at time t: From all
Sequential Monte Carlo Simulation. The Latin Hypercube the available conventional generating units, the WTG, and
Sampling (LHS) can effectively improve the efficiency of CHPP are combined to create the total system generating
sampling and space coverage with small amount of calculation
capacity. Then, compare the available generation capacity
compared to random sampling.
with the total system load to identify generation deficiencies
The LHS method is used to generate a series of up or on an hourly basis.
down times series, called state residence time series, for each a) If load shedding occurs: enter into releasing extra
component in the system. And then, the series of all the water. The extra discharge water flow of ith hydroplant is
components are combined together as the system status series.
proportional to its normal capacity Ci, which is expreesed by
(15), Ait is the operating state of ith CHPP.
⎛ k − 0.5 ⎞
xnk = Fn−1 ⎜
⎝ K ⎠
⎟ (13)
(
Qib = min Qimax − Qit , Qib _ max1 , (Vi t −1 − Vi min ) 3600 ) (15)
Where the is the extra released water flow of ith
Qib_max1
Where Fn is the cumulative probability function of the nth
CHPP theoretically, which can be obtained by:
variable xn. The LHS can be used sample the states duration of
generations, the wind speed, water inflow and the load. ⎛ I
⎞
Qib _ max1 =-EENS ⎜ Ci Ait ηi ∑ Ci Ait ⎟ (16)
B. Assessment Indices ⎝ i =1 ⎠
The most widely used reliability index is the Expected b) If load shedding does not occur: extra water may be
Energy Not Served (EENS). Load shedding calculation is entered into the revior of CHPP, which can be calculated as:
required at every time sequence in order to obtain this index.
 (
Qib = min Qit − Qimin , Qib _ max 2 , (Vi t −1 − Vi max ) 3600 (17) ) The correlation of the four wind farm can be found in table
II. Due to the limited historical data, it is assumed that location
Where the Qib_max2 is the extra reserved water flow of ith of the wind farms in the northwest Grid can be represented by
CHPP theoretically, which can be obtained by: these four wind farms. The Weibull parameters are obtained
⎛ I
⎞ by the parameter fitting of historical wind speed data. The cut-
Qib _ max 2 =EENS ⎜ Ci Ait ηi ∑ Ci Ait ⎟ (18) in, rated and cutout wind speeds of the Wind turbine unit are 3,
⎝ i =1 ⎠ 10.5, and 20m/s, respectively, each wind turbine is rated at
The contribution of the available extra power from the 1.5MW. The 7 cascaded hydropower plants information can
reservoirs of CHPPs to the variation of EENS can be be found in table III. The lower and upper limit of water
calculated by: volume is 0.4 and 1 of normal water volume, respectively.
I The mean monthly water inflow ratio of the yearly mean
Δeens = ∑ηi Q 3600Δt i
b
(19) water inflow can be found in Fig.1. The normal distribution is
i =1
assumed, the standard deviations of the water inflow are
assumed as 6.6% and 3.3% of the mean value in the dry
The initial volume and the minimum of the reservoir are season and rainy seasons, respectively.
assumed to be at 80% and 30% of its maximum volume,
respectively. From the viewpoint of power system, EENS is TABLE II. CORRELATION BETWEEN MULTI-WIND FARM
reduced due to the coordination of wind-hydro plants, the
CHPP can be used for emergency dispatch under power Wind farms
system failure conditions. Wind farms South
ChangMa LiuYuan
MaZong
Ganhekou Mountain
5) Proceed to the next time interval: Repeat (2)–(5) until South
1 0.435 0.730 0.195
Ganhekou
it reaches the end of year (i.e., the 8736 or 8760th hour).
ChangMa 0.435 1 0.431 0.463
6) Check whether the EENS index is converging: the LiuYuan 0.730 0.431 1 0.307
coefficient of variation is used : MaZong
0.195 0.463 0.307 1
σ Mountain
β = EENS (20)
μ EENS TABLE III. CASCADED HYDROPOWER PLANTS
Y
σ EENS = ∑ ( EENS − uEENS ) (Y (Y − 1) ) (21)
2 Cascaded hydropower plants
y Parameter
H1 H2 H3 H4 H5 H6 H7
y =1
Y Mean water
647 659 664 717 799 994 546
μ EENS = ∑ EENS y Y (22)
inflow (m/s)
Vimax(108m3) 247 11 16.5 6.2 2.64 57 25.8
y =1 Pimin(MW) 590 990 581 492 332.3 400 175
Where β is the coefficient of variation, σEENS and μEENS are the Ci (MW) 1280 3500 1600 1500 1020 1350 800
standard deviation and the expected estimate of the EENS, Qimin (m/s) 552 536 553 536 705 488 333
Qimax (m/s) 1198 1896 1522 1635 2166 1650 1522
respectively. Y is the number of simulated years.
ηi (10-4 kWh/m3) 2.97 5.13 2.92 2.55 1.31 2.27 1.46
Repeat (1)–(6) until the coefficient of variation is less than
the specified tolerance error.

V. CASE STUDY
A. Northwest Grid Test System Data
The equivalent of provincial power system in northwest
China was used in this study because it contains a large
number of CHPPs with high penetration of wind generation.
The equivalent system load is 26353.9MW, of which
5000MW is for exporting to areas outside the province. The
detailed information can be found in table I. There are 20 wind
farms in this grid, 20 thermal units, and 7 CHPP.
Figure 1. The mean month water inflow
TABLE I. POWER SUPPLY AND DEMAND
B. Reliability Assessment Results
Generation Data(MW) Due to the historical data characteristic, the 8736 hours is
Load(MW)
Thermal hydro Wind used. The independent mode is chosen as a comparative case
so that the cascaded hydro plants don’t participate in the
26353.9 2380 11050 3508.5
dispatch. The power output of cascaded hydro plants is
determined by the water inflow and power output limits. The
independent system reliability is also assessed. The impact of Table V shows the impact of the number of hydro units
coordination of the generating units on system adequacy is assigned to coordinate with wind power on the system LOLE,
investigated. From Fig.2, it can be found that the LHS can LOLF, and EENS, which all decreased as the number of
effectively reduce the amount of calculation. Through 2000 coordinated hydro units’ reservoir increases from 1 to 7. It can
years simulation, the coefficient of variation become stable, be seen that the system reliability indices falls as the
and the simulation results should be persuasive. coordinated water reservoir volume rises.

VI. CONCLIUSION

This paper presented results of power system reliability

assessment considering the impact of the coordination of wind
generation together with CHPPs. The characteristics of the
time series of the variable energy sources of wind farms and
cascade hydropower, and load were preserved by the use of
time sequential Monte Carlo simulation based on Latin
Hypercube Sampling. The correlation between multi-wind
farms is considered. The proposed method is applied to the
northwest grid of China. Results suggest that through the
coordination of wind and hydropower the system reliability
Figure 2. Coefficient of variation
can be improved significantly.

TABLE IV. RELIABILITY INDICES REFERENCES

[1] J. Wen, Y. Zheng, F. Donghan, “A review on reliability assessment for
Operation mode wind power,” Renewable and Sustainable Energy Reviews, vol.13, no.9,
Indices
Independent Coordination pp.2485–2494, Dec. 2009.
[2] R. Billinton, Y. Gao, and R. Karki, “Composite system adequacy
EENS (MWh) -41,753 -8398 assessment incorporating large-scale wind energy conversion systems
considering wind speed correlation,” IEEE Trans. Power Syst., vol. 24,
LOLF (h) 15.18 6.51 no. 3, pp.1375–1382, Aug. 2009.
[3] A. E. Feijoo, J. Cidrab, and J. L. G. Dornelas, “Wind speed simulation
LOLP (%) 0.1737 0.0745 in wind farms for steady-state security assessment of electrical power
systems,” IEEE Trans. Energy Convers., vol. 14, no. 4, pp.1582-1588,
Besides EENS, the Loss of load frequency (LOLF) and Dec. 1999.
loss of load probability (LOLP) are used to assess the system [4] W. Wangdee and R. Billinton, “Considering load-carrying capability
and wind speed correlation of WECS in generation adequacy
reliability. These indices can be found in Table IV. The
assessment,” IEEE Trans. Energy Convers., vol. 21, no. 3, pp.734-741,
reliability indices are decreased significantly. EENS is Sep. 2006.
decreased to 8398MWh/year. The LOLF and LOLP are [5] F. Port´e-Age, Y.T. Wu and C. H. Chen, “A numerical study of the
decreased by 57.11%. It can be seen that the coordination of effects of wind direction on turbine wakes and power losses in a large
wind-hydro can improve the reliability of power system, and wind farm,” Energies, vol. 6, pp. 5297-5313, Oct. 2013.
the reliability model is also verified. [6] H. Kim, C. Singh, and A. Sprintson, “Simulation and estimation of
reliability in a wind farm considering the wake effect,” IEEE Trans.
Sustainable Energy, vol. 3, no. 2, pp.274-282, April, 2012.
C. Sensitivity on the Usable Storage [7] C. González, J. Juan, J. Mira, F. J. Prieto, and M. J. Sánchez,
The effect of usable reservoir volume on the system “Reliability analysis for systems with large hydro resources in a
reliability is analyzed in this paper by considering the deregulated electric power market,” IEEE Trans. Power Syst., vol. 20,
no. 1, pp. 90-95, Feb. 2005.
different number of hydro units taking part in coordinated [8] V. S. Lopes and C. L. T. Borges, “Impact of the combined integration
operation with wind power. of wind generation and small hydropower plants on the system
reliability,” IEEE Trans. Sust. Energy, in press.
TABLE V. THE RELIABILITY INDICES TREND AS THE NUMBER OF [9] R. Karki, P. Hu and R. Billinton, "Reliability assessment of a wind
COORDINATED CASCADE HYDROPOWER PLANT integrated hydro thermal power system," in Proc. 2010 IEEE 11th
International Conference on Probabilistic Methods Applied to Power
Number of Systems (PMAPS), pp. 265-270.
coordinated LOLF (h) LOLP (%) EENS (MWh) [10] R. Karki, P. Hu and R. Billinton, “Reliability Evaluation Considering
hydro units Wind and hydro power coordination," IEEE Trans. Power Syst., vol. 25,
1 13.349 0.15% -35817 no.2, pp.685-693, May, 2010.
2 10.834 0.12% -22517 [11] L. Y. Chen and Y. Liu, "Scheduling strategy of hybrid wind-
3 9.902 0.11% -18149 photovoltaic-hydro power generation system," in Proc. International
4 8.894 0.10% -14522 Sustainable Power Generation and Supply Conf., pp.1-6, 2012.
5 7.681 0.0879% -12335 [12] H. J. Chen, J. X. Wang, Y. Zhang. "Economic dispatch of hydro-
6 6.616 0.0757% -9871 thermal power system with large-scale wind power penetration," in
7 6.512 0.0745% -8398 Proc. 2012 IEEE Asia-Pacific Power and Energy Engineering Conf.
pp. 1-4.