www.ietdl.

org
Published in IET Generation, Transmission & Distribution
Received on 4th October 2012
Revised on 3rd September 2013
Accepted on 11th September 2013
doi: 10.1049/iet-gtd.2013.0340

ISSN 1751-8687

Evaluation of nodal reliability risk in a deregulated

power system with photovoltaic power penetration
Qian Zhao1, Peng Wang1, Lalit Goel1, Yi Ding2
1
School of Electrical and Electronic Engineering, Nanyang Technological University (NTU), 639798 Singapore, Singapore
2
Department of Electrical Engineering, Technical University of Denmark, Lyngby, Denmark
E-mail: yding@elektro.dtu.dk

Abstract: Owing to the intermittent characteristic of solar radiation, power system reliability may be affected with high
photovoltaic (PV) power penetration. To reduce large variation of PV power, additional system balancing reserve would be
needed. In deregulated power systems, deployment of reserves and customer reliability requirements are correlated with
energy and reserve prices. Therefore a new method should be developed to evaluate the impacts of PV power on customer
reliability and system reserve deployment in the new environment. In this study, a method based on the pseudo-sequential
Monte Carlo simulation technique has been proposed to evaluate the reserve deployment and customers’ nodal reliability with
high PV power penetration. The proposed method can effectively model the chronological aspects and stochastic
characteristics of PV power and system operation with high computation efficiency. An auto-regressive and moving average
model has also been developed for simulating the chronological characteristics of the solar radiation. Customers’ reliability
preferences have been considered in the generation and reserve deployment. Moreover, the correlation between PV power and
load has been considered in the proposed method. Nodal reliability indices and reserve deployment have been evaluated by
applying the proposed method to the Institute of Electrical and Electronics Engineers reliability test system.

Nomenclature uji , ujk voltage phased angle at bus i, k, respectively

max

S
maximum apparent power on transmission line
0 normal state index (superscript) ik
j contingency state index (superscript) from bus i to k
0
i, k bus index (superscript) Pig real power output of unit g at node i at normal
g generator index (subscript) state
j
d load sector index (subscript) DPig primary reserve real power provided by unit g at
N set of buses node i at state j
Ng set of generator buses Rjig second class real power reserve provided by unit
Nd set of load buses g at node i at state j
0
NGi set of generators Pid normal state load of load sector d at node i
j
NDi set of load sectors DPid load curtailment of load sector d at node i at state j
NPV set of load buses containing PV power min
Pig minimum real power output of unit g
P̂PVi predicted PV power at node i max
Pig maximum real power output of unit g
0
PPVi actual PV power at node i at normal state Rmax maximum second class reserve power provided
j ig
PPVi actual PV power at node i at state j by unit g
0
P̂NLi predicted net load (NL) at state 0
0
PNLi actual NL at state 0
j
PNLi actual NL at node i at state j 1 Introduction
Cig energy cost function
CRig secondary real power reserve cost function In recent years, renewable energy sources have been rapidly
Cid load interruption cost function adopted in many countries and regions as substitutes for
P jG generation power vector traditional energy resources [1] for reducing environmental
pollutants and global warming effects. Renewable energy
P jNL NL power vector
has also been gradually competitive with conventional
Bj admittance matrix of transmission network energy in markets because of the rising costs of
Θj bus voltage phase angle vector conventional energy and government tariff policies [2].
xjik line impedance from bus i to k Solar energy is a promising renewable energy source

IET Gener. Transm. Distrib., 2014, Vol. 8, Iss. 3, pp. 421–430 421
doi: 10.1049/iet-gtd.2013.0340 & The Institution of Engineering and Technology 2014
www.ietdl.org
because of its abundance, accessibility and clean generation
process. The application of photovoltaic (PV) panels for
generating electric power also promotes the utilisation of
solar energy [1].
However, the power output of a PV generating system
comprising a number of PV panels is mainly determined by
the available solar radiations (SRs) which vary
chronologically. The fast fluctuation of the SRs makes the
power output of the PV generating system intermittent and
totally different from that of the conventional generators.
The PV power is usually high around noon and not
available during the night. To utilise solar power on a
large-scale effectively and efficiently, its chronological and
intermittent characteristics have to be modelled [3–6].
The high PV power penetration in electric power systems
can impact the generation and reserves deployment, and
therefore bring complexities in power system reliability [4–
12]. In power system operation, operating reserve and
contingency reserve are dispatched to meet load variation
and capacity inadequacy in contingency state, respectively,
[7, 12]. Both types of reserve can be provided by spinning
reserve, online generating units or fast start-up generating
units with different synchronous times. With high
renewable energy penetration, renewable power fluctuation
may impact utilisation of both types of reserves, and the
differentiation between them is obscure [7]. It is illustrated
in [8] that adequate reserve is necessary for balancing the
high fluctuation of renewable power output. Some research
has shown that the requirement of reserve will ascend with
the increase of wind power penetration for maintaining
power system reliability levels [5, 9]. However, the reserve
requirement in the new environment also depends on the
correlation between renewable energy and load. If their
correlation is positive, less reserve may be needed and vice
versa. The prediction errors of renewable power output can
also impact the reserve deployment in deregulated power
systems [10], where customers are more concerned about
their own reliabilities than system reliabilities and also have
their reliability preferences, which are evaluated by
customers’ nodal reliability indices [13]. The utilisation of
PV power will be affected by customers’ reliability
preferences. Therefore it is important to develop a method
for evaluating customers’ nodal reliabilities that can take
into consideration customers’ reliability preferences and
high power fluctuation caused by PV penetration.
The existing methods for power system reliability
evaluation can be classified into two categories, analytical
methods and Monte Carlo simulation methods [14–16]. The
analytical methods calculate system reliability indices for
determined states using direct mathematical methods.
Monte Carlo simulation methods, on the other hand,
simulate the stochastic behaviour of the power system
operation for estimating reliability indices [14]. The
analytical methods are relatively easy to implement.
However, they are very sensitive to the system size, which
usually takes significant computational efforts for evaluating
reliabilities of large systems with complex generating
conditions. On the other hand, when considering complex
operating conditions or when the number of contingency
events is large, Monte Carlo simulation is preferred as the
number of samples required for a given accuracy level is
independent of the system size [17]. Monte Carlo
simulation methods can be further sub-divided into
sequential and non-sequential methods. Sequential Monte
Carlo simulation represents modelling the chronological
aspects of system operation. However, it takes greater
422
& The Institution of Engineering and Technology 2014
computation effort and sometimes a solution is impossible
especially for large-scale power systems. Non-sequential
Monte Carlo simulation, on the other hand, takes less
computational effort, but cannot model the chronological
characteristics of power system operation. The
pseudo-sequential Monte Carlo simulation preserves the
high computation efficiency of the non-sequential Monte
Carlo simulation and the chronological property of the
sequential Monte Carlo simulation [18–20]. The
pseudo-sequential Monte Carlo simulation techniques were
initially developed for conventional power systems [18–20]
and have been extended in this paper for reliability
evaluation of deregulated power systems with high PV
power penetration.
This paper proposes a new method for evaluating the
impact of high PV power penetration on customers’ nodal
reliability and system energy and reserve deployment. The
new indices for assessing system reserve commitment and
deployment have been developed through the proposed
method. The auto-regressive and moving average (ARMA)
model has been utilised to simulate the chronological
characteristics of the SR. Pseudo-sequential Monte Carlo
simulation has been conducted in the proposed method for
taking into account both chronological and stochastic
characteristics of PV power as well as customer reliability
preferences. The proposed method has also considered the
correlation between PV power and load. Customers’ nodal
reliability and reserve deployment of generating units have
been formulated and evaluated, respectively. The modified
Institute of Electrical and Electronics Engineers (IEEE)
reliability test system (RTS) [21] is used to illustrate the
method. The modelling of PV generating system has been
discussed in Section 2. In Section 3, optimal energy and
reserve scheduling formulation for power system operation
considering PV power penetration has been introduced. The
proposed method based on the pseudo-sequential Monte
Carlo simulation for nodal reliability and reserve evaluation
has been introduced in Section 4. The system studies based
on the modified RTS are given in Section 5.
2 PV generating system modelling
In this section, a PV generating system has been modelled and
analysed in terms of its power output, reliability and
penetration.
2.1 Solar radiation
PV power output is mainly determined by SR, which is
affected by the weather at a specific location. Clearness
index is defined as the ratio of the average horizontal SR at
the site to the extraterrestrial radiation on a horizontal
surface above the site just outside the atmosphere [22]. The
clearness index can reflect both the environmental and
meteorological factors. This index is therefore widely used
for determining SR [22–24]. An ARMA model of clearness
index is developed using the SR data from Singapore.
At hour t, the indices for SR are defined as: OSRt is the
observation of SR at hour t; CSRt is the clear day SR at
hour t; μt is the mean value of OSRt; and σt is the standard
deviation of OSRt.
CSRt is determined using American Society of Heating,
Refrigerating and Air Conditioning Engineers (ASHRAE)
model instead of the extraterrestrial radiation [22]. The
hourly clearness index kt and mean clearness index mkt for
IET Gener. Transm. Distrib., 2014, Vol. 8, Iss. 3, pp. 421–430
doi: 10.1049/iet-gtd.2013.0340
 www.ietdl.org
hour t, and the daily clearness index for day n, dkn are Under SR and temperature T, the maximum output power
determined using the following equations is calculated as

k t = OSRt /CSRt (1) T , SR

mk t = mt /CSRt
 
dkn = OSRt / CSRt
t[n t[n
For removing the diurnal effects, the hourly clearness index kt
is normalised by subtracting mkt and then divided by the
standard deviation. The normalised data series yt and the
ARMA model of yt are evaluated in (4) and (5), respectively
k t − mk t
yt =
st
yt = f1 yt−1 + f2 yt−2 + · · · + fp yt−p
+ at − u1 at−1 − u2 at−2 − · · · − uq at−q
where φi (i = 1, 2, …, p) and θj ( j = 1, 2, …, q) are the
auto-regressive and moving average parameters of the
model, respectively. αt is a normal white noise process
 with
zero mean and a variance of s2a (i.e. at [ NID 0, s2a ,
where NID denotes normally independently distributed).
After obtaining the predicted series of yt from (5), the
predicted clearness index skt and simulated solar radiation
SSRt at time t can be evaluated by using (6) and (7),
respectively
Pmp = IscT , SR Voc
T , SR
ff (8)
(2)
where IscT , SR and VocT , SR
are the short-circuit current and
(3) open-circuit voltage of the PV panel, respectively, which
can be evaluated by the procedures proposed in [25]. Fill
factor ff is usually assumed to be a constant value, which
measures the ratio of the maximum power Pmp to the
product of IscT , SR and VocT , SR
.
The electric parameters of the PV panel utilised in this
paper are shown in Appendix 1. For a specific site, the
relationship between the power output and the SR is
obtained using (8) from the measurement SR. A piecewise
(4) linear function is obtained for the power output curve using
the least-square fitting method. Fig. 2 shows the
relationship between the PV output and the SR. From the
(5) least-square fitting curve, the maximum PV panel power
output Pmp and the PV array output PPV are determined by
(9) and (10), respectively
Pmp (SR) = 0.2548 SR + 0.0027 kW (9)
PPV = Pmp (SR) × NPV (10)

sk t = yt st dkn + mk t (6)
SSRt = sk t CSRt (7)

2.2 PV array power output

The power output of PV panel is determined by I–V and P–V
curves of the panel. Fig. 1 illustrates the I − V curves for
different cell temperature and SR levels. The mpp in the
figure denotes the maximum power point of the PV panel.
The Impp and Vmpp represent the current and voltage at the
mpp, respectively. The parameters of the two curves are
mainly affected by the cell temperature and radiation level
[22, 25]. Fig. 2 Relation between PV panel output power and SR

Fig. 1 PV panel output curves

IET Gener. Transm. Distrib., 2014, Vol. 8, Iss. 3, pp. 421–430 423
doi: 10.1049/iet-gtd.2013.0340 & The Institution of Engineering and Technology 2014
www.ietdl.org
where SR is NPV are the SR and number of PV panels for the
PV array, respectively.
It can be observed from (9) that the power output of a PV
panel is approximated as a linear function of SR – the ratio of
power output of the panel against the SR is: Pmp (SR)/SR ≃
0.2548 kW/panel. Since PV panel surface area is about
1.65 m2 as shown in Appendix 1, for 1 kW/m2 SR, the
power output (kW)/m2 of the panel is: 0.2548 kW/1.65 m2
= 0.152 kW/m2. Therefore the conversion efficiency of the
panel is 15.2%.
2.3 Forecast error of PV array power output
The actual and predicated power output of a PV array are
denoted as PPV and P̂PV , respectively. The prediction error
ΔPPV is represented as
DPPV = PPV − P̂PV (11)
The predicted power output and prediction error of a PV array
with 1 × 105 PV panels were examined in this paper. The
probability density function (PDF) of the PV array power
output from predicted SR and measurement are shown in
Fig. 3. The fit of a t-distribution with freedom 1 to the PV
array power output prediction errors is shown in Fig. 4.
2.4 PV array reliability
PV arrays are usually distributed at load points. In reliability
analysis, a PV array is modelled as a two-state Markov model.
The failure and repair rates used are 0.000383 (1/h) and
0.00381 (1/h), respectively [26, 27].
2.5 PV power penetration
A PV array located at the demand side will be utilised for
supplying load directly. In addition, from an economic
point of view, PV power should be consumed first. It is
assumed that customers will not use energy storage for
storing PV power to make energy arbitrage. The load that
cannot be fulfilled by PV energy is the NL, which is
supplied by the conventional generation.
Fig. 3 PDF of PV array power output
424
& The Institution of Engineering and Technology 2014
Fig. 4 PDF of PV array power output prediction error
0 0
Predicted NL P̂NLi and actual NL PNLi at node i is
determined by (12) and (13), respectively

0
P̂NLi = 0
Pid − P̂PVi (12)
d[NDi

0
PNLi = 0
Pid − PPVi
0
(13)
d[NDi
3 Problem formulation
PV power is utilised in power system operation to substitute
part of the conventional energy. However, the unpredicted
PV power fluctuation needs to be balanced by system
reserves within an acceptable time period (e.g. 10 min or
intra-hour). PV power fluctuations may also require
utilisation of down reserves, which is seldom used in a
conventional power system with low penetration of
renewable energy resources [12]. Most electricity markets
would provide balancing reserves with different
synchronous times or various utilisation purposes [28]. For
example, in the Nordic power system, the frequency
controlled normal operating reserve and frequency
controlled disturbance reserve are procured to restore energy
balances caused by demand variation and system
disturbances, respectively [28]. In this paper, balancing
reserves are modelled as primary reserve and secondary
reserve. Primary reserve can be up reserve or down reserve
and is mainly used to meet the unpredicted load variation
or unpredicted PV power. Secondary reserve is the up
reserve and is mainly used for contingency events when
large generation inadequacy occurs. In deregulated power
systems, customers may choose to interrupt their electricity
consumption if the energy cost is higher than their
reliability benefits. Customers’ reliability benefits are
difficult to be measured directly and often evaluated
indirectly by their interruption costs [14]. Therefore
customers’ interruption costs should be considered in the
energy and reserve scheduling models.
The optimal energy and reserve scheduling problem in
power system operation is formulated in two steps as shown
in [7]. In the first step, the unit commitment (UC) is
scheduled for each operating hour. The algorithm for
determining UC is obtained from [29]. Each generating unit
provides the cost functions for both energy and reserve to
the system operator reflected by the corresponding bids.
IET Gener. Transm. Distrib., 2014, Vol. 8, Iss. 3, pp. 421–430
doi: 10.1049/iet-gtd.2013.0340
 www.ietdl.org
The units at the least operation cost are committed for Transmission line constraints
generation, and reserves will also be determined for


satisfying the ‘N − 1’ criterion, which is important
1 


max

j j 

j

for maintaining the reliable operation of power systems. In
j ui − uk
≤
Sik
(18)

xik

the evaluation, different generating units can have different
ramping capabilities. However, the generating units for
providing reserves must satisfy the criteria determined by
the system operators. For example, reserves must be fully
4 Proposed method for nodal reliability and
activated in 15 min in Nordic power system. reserve evaluation
In the second step, the reserve utilisation and load
4.1 Markov models of system components
curtailment are determined for each contingency state
during the real-time operation. When a power system The reliability models of generating units, transmission lines
transfers from the normal operating state to a contingency and PV arrays are represented as two-state Markov models
state because of random failures, prices may change with [14]. The failure rate λ of and the repair rate μ of a system
system operating condition. Some customers may interrupt component are evaluated by (19) and (20), respectively
their loads when electricity prices increase dramatically.
Customer response to the change of prices associated with 1  
the interruption can be indirectly measured by the customer l= 1/h (19)
MTTF
interruption cost, which represents the customer willingness
to pay to avoid service interruption [13]. In other words, the 1  
customer will bid a price, which reflects the customer m= 1/h (20)
MTTR
willingness to pay to avoid service interruption [30]. Sector
customer damage function (CDFs) [14] estimating the where MTTF and MTTR are the mean time to failure and
different customer sectors interruption cost is used to model mean time to repair, respectively.
customer response to load curtailment [13, 30]. The input The availability of a component pin and unavailability pout
data of CDFs which is important for cost analysis has been are determined by (21) and (22), respectively
provided in Table 1 in Section 5.
A DC optimal power flow (OPF) model has been m
pin = (21)
developed, which is computationally efficient. The objective l+m
of the OPF model is to minimise the generation, reserve
l
utilisation and customer interruption cost for system state j pout = (22)
l+m
    j 
min fj = Cig DPig + Pig
0
j j
0 , DP , R , DP j
Pig ig ig id j
i[Ng g[NGi
j
4.2 Pseudo-sequential Monte Carlo simulation
     
+ CRig Rjig + j
Cid DPid (14) In the pseudo-sequential Monte Carlo simulation process, a
i[Nd d[NDi non-sequential Monte Carlo simulation method is used to
select the load level and system state. If the selected state is
subject to the following constraints: a failure state, a sequence of neighbouring states and the
DC power balance equation duration of failure are determined utilising the state
transition method [18, 19]. The pseudo-sequential
Bj Qj − P jG + P jNT = 0 (15) simulation process for evaluating customers’ nodal
reliability and reserve deployment considering PV power
where the ith element of P jG is the generation power at node i fluctuations is summarised in the following steps:
 j
for state j and equals g[NGj (Pig 0
+ DPig + Rjig ), and the ith
i Step 1: generate the hourly time varying sequences of PV
j
element
 of 0P NL is j the
 NLj at node i for state j and equals power output of PV generating systems. Calculate predicted
d[NDi P id − DP id − PPVi . NL profile using (12).
0
Generating unit limits Step 2: sample a predicted NL level P̂NLi from the predicted
NL profile based on a uniform distribution.
j
min
Pig ≤ DPig + Pig
0
≤ Pig
max
(16) Determine the committed units and their scheduled power
0
outputs Pig for the operating hour based on the UC, which is
the first step of optimal energy and reserve scheduling
0 ≤ Rjig ≤ Rmax
ig (17) illustrated in Section 3.

Table 1 Parameter values for CDFs of different sectors [14]

Failure durations, min Interruption costs, $/kWh

Large users Industrial Commercial Agricultural Residential

1 0.073 0.46 0.129 0.027 0.0004

20 0.111 1.332 1.014 0.155 0.044
60 0.163 2.99 2.951 0.295 0.243
240 0.291 8.899 10.992 1.027 2.235
480 0.604 18.156 28.02 2.134 6.778

IET Gener. Transm. Distrib., 2014, Vol. 8, Iss. 3, pp. 421–430 425
doi: 10.1049/iet-gtd.2013.0340 & The Institution of Engineering and Technology 2014
www.ietdl.org
Step 3: sample a system state j by sampling the states of where
committed units, transmission lines and PV generating

systems based on their respective probability density
 0 j = success state
function, which is generated using the method presented in EICj = Kd /E(D) j = failure state
Section 4.1. d[NLj
Step 4: evaluate the system state j: if it is a successful state (all
the components are in their successful state), go to step 2; if it and Kd is the interruption cost for customer sector d, NLj is the
is a failure state, go to the next step. set of customer sectors. Kd is evaluated as
Step 5: if the PV generating system failed at node i, the load at 
node i will be Pid0
. Otherwise the load is PNLi 0
(NL) at node i,  
Kd = CDFd Dk × DPid
k
(26)
which considers the power contribution of the PV generating k[I j
system.
Step 6: calculate the failure states sequence starting from state k is one of the state of the failure sequence I j for selecting
j utilising the forward/backward simulation. failure state j. CDFd(Dk) is the CDF of customer sector d
Step 7: calculate the failure state duration Dj according to the for duration Dk, E(D) is the expected duration for the failure
failure sequence obtained in step 6. sequence I j.
Step 8: solve the optimisation problem proposed in (14)–(18) The expected committed primary (ECP) reserve
for determining the reserve deployments and load curtailment
for state j, which is the second step of optimal energy and 
N
reserve scheduling illustrated in Section 3. ECP = CPRj /N (27)
Step 9: calculate customers’ nodal reliability and reserve j
indices for state j.
Step 10: go to step 3 if convergence criteria is not satisfied, where
otherwise go to step 11.
Step 11: calculate nodal reliability and reserve indices based ⎧
⎨ 0 j = success state
on the equations described in Section 4.3.    
The method for generating forward/backward state CPR =j max
Pig − 0
Pig ×T j = failure state
⎩ i[N j
sequences and failure duration E(D) in steps 6 and 7 are g g[NGi
explained in Appendix 2 [18, 19].
The expected committed secondary (ECS) reserve
4.3 Nodal reliability and reserve indices

N

The expected energy not supplied (EENS) is the estimator of ECS = CSRj /N (28)
j
energy not supplied (ENS)
where

N
EENS = ENS /N j
(23) ⎧
j=1 ⎨  0 j = success state
CSRj = Rmax × T j = failure state
⎩ i[Ng g[NGj ig
where i

⎧ The expected utilisation of primary (EUP) reserve

⎨   0 j = success state
j
ENSj = DPNLi ×T j = failure state
⎩ i[Nd d[NLj 
N
i EUP = PRj /N (29)
j
where T is the examination period, for example, 8760 h for
annual analysis. where
The loss of load probability (LOLP)
⎧
⎨ 0 j = success state
    
N PR =
j j
Pig − 0
Pig ×T j = failure state
LOLP = F j /N (24) ⎩ i[N j
g g[NGi
j=1

The expected utilisation of second class (EUS) reserve

where
 
N
0 j = success state EUS = SRj /N (30)
F =
j
1 j = failure state j

The expected energy interruption cost (EEIC) where

⎧

N ⎨  0 j = success state
EEIC = EIC /N
j
(25) SRj = Rj × T j = failure state
⎩ i[Ng g[NGj ig
j=1 i

426 IET Gener. Transm. Distrib., 2014, Vol. 8, Iss. 3, pp. 421–430
& The Institution of Engineering and Technology 2014 doi: 10.1049/iet-gtd.2013.0340
 www.ietdl.org
The uncertainty of EENS is evaluated by its variance Table 2 Nodal PV array installed capacity, MW
coefficient. The coefficient of variance of EENS is used as
Nodes Case 1 (9.8% PV Case 2 (19.6% PV
the convergence criterion in step 10 of the penetration) penetration)
pseudo-sequential Monte Carlo simulation described in
Section 4.2 1, 2, 3, 4, 5, 6, 7, 28 28
8, 9 and 10
 13, 15 and 18 0 56
bEENS = V (EENS)/EENS (31) 14, 16, 19 and 20 0 28

where V(EENS) = V(ENS)/N is the variance of EENS,

V(ENS) is the variance of ENS.
Table 3 EENS, MWh/year
5 Case studies Nodes Case 0 Case 1 (a) Case 1 (b) Case 2 (a) Case 2 (b)

The modified IEEE RTS [21] is used to illustrate the proposed 1 2446.8 3120.9 2263.1 4936.3 2230.4
techniques. The single line diagram of RTS is shown in 2 3041.8 3934.0 2789.0 6388.3 2774.8
Fig. 5. The customer interruption cost for different 3 3048.8 3937.2 2830.0 6224.5 2775.3
4 2380.0 3016.5 2207.3 4708.2 2161.2
interruption durations is obtained from the CDFs and shown 5 2090.5 2648.1 1951.6 4081.3 1897.0
in Table 1 [14]. Three PV penetration levels of 0% (case 0), 6 1943.5 2443.2 1821.8 3707.6 1761.2
9.8% (case 1) and 19.6% (case 2) are studied. The nodal 7 2348.2 2932.4 2196.1 4865.7 2161.1
penetrations of PV power are shown in Table 2. To 8 3986.7 5261.0 3709.9 8611.7 3651.8
9 3217.7 4171.2 2983.4 6642.8 2928.0
quantify the impact of PV power with or without additional 10 2903.0 3727.7 2684.7 5883.0 2626.8
reserve, the simulations for cases 1 and 2 have been done 13 3589.1 4815.6 3339.7 7637.5 3249.7
under two conditions. Under condition (a), the UC of 14 3250.4 4289.9 3062.8 6769.8 2981.0
conventional units is determined based on the NL, which is 15 3460.2 4773.2 3315.7 7384.9 3185.4
the load minus the PV power output. Under condition (b), 16 3142.7 4314.3 2987.9 6761.2 2885.3
18 3874.3 5301.4 3740.4 7883.0 3611.4
the UC for conventional units are only determined based on 19 4173.6 5847.4 4001.3 9210.9 3860.8
0
Pid and PV powers are treated as additional power. 20 2506.4 3338.6 2364.7 5266.6 2298.1
The EENS and LOLP at different load nodes are shown in
Table 3 and Fig. 6, respectively. Under condition (a),
customers’ reliabilities are impaired by PV power
fluctuations and unreliability of PV generating systems.
EENS and LOLP increase as PV power penetration

Fig. 6 LOLPs at different load nodes

increases. For customers at node 1–4, 7–20, EENS increase

more than 100% from case 0 to case 2 (a), whereas for
customers at nodes 5–6, EENS increase less than 100%.
Customers’ LOLP also have different variations after PV
penetration. For condition (b), EENS and LOLP decrease
after PV penetration increases since PV power served as an
additional power source. Improvements of customers’
EENS and LOLP are not uniform for each node under
condition (b). For customers at nodes 13, 15 and 18 with
higher PV power penetration, reliability improvement is less
than other nodes.
Therefore in deregulated power systems, customers’
reliability may be compromised by the PV penetration if
there is insufficient reserve to account for the unpredicted
PV power fluctuations. The impacts of PV power
penetration on customers’ nodal reliability also depend on
Fig. 5 IEEE RTS [21] customers’ choices and should be evaluated after

IET Gener. Transm. Distrib., 2014, Vol. 8, Iss. 3, pp. 421–430 427
doi: 10.1049/iet-gtd.2013.0340 & The Institution of Engineering and Technology 2014
www.ietdl.org
considering customers’ reliability preferences. In an sources in power system operations. However, under such
interconnected power system with high PV penetration, for conditions, the reserve deployment and utilisations will
customers located at nodes where PV power is penetrated, increase greatly.
the variations of nodal reliability are not more significant
than for other customers.
The EEIC is shown in Fig. 7. Customers’ EEIC increases as
6 Conclusions
PV penetration increases. EEIC increase by more than 400% In this paper, nodal reliability and reserve deployment have
from case 0 to case 2 (a), whereas it increases less than 100% been studied for deregulated power systems with high PV
from case 0 to case 1 (a). We can observe that with high PV power penetration. A method utilising a pseudo-sequential
penetration, the unreliability of PV arrays may increase the Monte Carlo simulation technique has been developed,
duration of the contingency events of the system, and which can take into consideration the stochastic and
customers’ EEIC will increase. However for different chronological fluctuations of PV power as well as
customers, the increases in EEIC are not linear and may customers’ reliability preferences. The results from the case
increase dramatically when solar penetration increases. studies illustrate that nodal reliability is compromised with
The reserve commitment and utilisation are shown in high PV power penetration because of the intermittence of
Tables 4 and 5, respectively. From Table 4, ECP and ECS PV generation system. The impacts of the PV generation
increase after PV power penetration especially under also depend on customers’ reliability preferences in
condition (b). Since PV penetration increases the power deregulated power systems. Therefore different PV power
fluctuations, the EUP and EUS increase under condition (a) penetrations have non-uniform impacts on customers at
to balance the PV fluctuations. The reserve deployment is different nodes. For maintaining the customer reliability
also affected by customers’ reliability preferences. Without level, the power system should have sufficient fast response
sufficient primary reserve, customers are more likely to be reserve for handling the large power fluctuations caused by
interrupted because of the slow response of secondary high PV power penetration. The proposed technique
reserve. Under condition (b), the EUP increases and the provides a flexible method to evaluate these impacts on
EUS decreases. With sufficient primary reserve, customers customers’ nodal reliability in deregulated power systems.
can maintain or even improve their reliability. This technique may also be extended to incorporate other
Therefore in deregulated power systems, the benefits of intermittent renewable energy sources.
solar power penetration on customers’ reliability highly
depend on whether the reserve markets can provide
sufficiently fast response and relatively cheap reserve for 7 References
balancing the fluctuations of PV power. The results shown
in this paper demonstrate that the RTS can have 20% PV 1 2008 Solar Technologies Market Report, 2010. Available at http://www.
nrel.gov/analysis/pdfs/46025.pdf
power integration without affecting the customers’ 2 Bird, L., Wüstenhagen, R., Aabakken, J.: ‘A review of international
reliability if PV power is integrated as additional energy green power markets: recent experience, trends, and market drivers’,
Renew. Sust. Energy Rev., 2002, 6, pp. 513–536
3 Billinton, R., Karki, R.: ‘Capacity expansion of small isolated power
systems using PV and wind energy’, IEEE Trans. Power Syst., 2001,
16, pp. 892–897
4 Billinton, R., Karki, B., Karki, R., Ramakrishna, G.: ‘Unit commitment
risk analysis of wind integrated power systems’, IEEE Trans. Power
Syst., 2009, 24, pp. 930–939
5 Doherty, R., O’Malley, M.: ‘A new approach to quantify reserve
demand in systems with significant installed wind capacity’, IEEE
Trans. Power Syst., 2005, 20, pp. 587–595
6 Ding, Y., Wang, P., Goel, L., Loh, P.C., Wu, Q.: ‘Long-term reserve
expansion of power systems with high wind power penetration using
universal generating function methods’, IEEE Trans. Power Syst.,
2011, 26, pp. 766–774
7 Papavasiliou, A., Oren, S.S., O’Neill, R.P.: ‘Reserve requirements for
wind power integration: a scenario-based stochastic programming
framework’, IEEE Trans. Power Syst., 2011, 26, p. 2197
Fig. 7 EEIC at different load nodes 8 Lew, D., Piwko, D., Miller, N., Jordan, G., Clark, K., Freeman, L.: How
Do High Levels of Wind and Solar Impact the Grid? The Western Wind
and Solar Integration Study, 2010, Available at http://www.nrel.gov/
wind/systemsintegration/pdfs/2010/lew_wwsis_grid_impact.pdf
Table 4 Expected reserve commitments, MW 9 Holttinen, H.: ‘Impact of hourly wind power variations on the system
operation in the nordic countries’, Wind Energy, 2005, 8, pp. 197–218
Case 0 Case 1 (a) Case1 (b) Case 2 (a) Case 2 (b) 10 Fabbri, A., GomezSanRoman, T., RivierAbbad, J., MendezQuezada, V.:
‘Assessment of the cost associated with wind generation prediction
ECP 183.85 286.19 323.99 315.72 395.44 errors in a liberalized electricity market’, IEEE Trans. Power Syst.,
ECS 168.04 254.05 263.29 276.20 293.29 2005, 20, pp. 1440–1446
11 da Silva, A., Sales, W.S., da Fonseca Manso, L.A., Billinton, R.:
‘Long-term probabilistic evaluation of operating reserve requirements
with renewable sources’, IEEE Trans. Power Syst., 2010, 25,
pp. 106–116
Table 5 Expected reserve utilisation, MWh/year 12 Ela, E., Milligan, M., Kirby, B.: Operating Reserves and Variable
Generation, 2011. Available at http://xwww.nrel.gov/docs/fy11osti/
Case 0 Case 1(a) Case 1(b) Case 2(a) Case 1(b) 51978.pdf
13 Wang, P., Ding, Y., Xiao, Y.: ‘Technique to evaluate nodal reliability
EUP 811 717 966 989 1 033 254 1 152 291 1 328 220 indices and nodal prices of restructured power systems’, IEE Proc.
EUS 265 210 302 340 260 514 383 277 261 900 Gener. Transm. Distrib., 2005, 152, pp. 390–396
total 1 076 927 1 269 329 1 293 768 1 535 568 1 590 121 14 Billinton, R., Allan, R.N.: ‘Reliability evaluation of power systems’
(Plenum Press, New York, 1996, 2nd edn.)

428 IET Gener. Transm. Distrib., 2014, Vol. 8, Iss. 3, pp. 421–430
& The Institution of Engineering and Technology 2014 doi: 10.1049/iet-gtd.2013.0340
 www.ietdl.org
15 Ding, Y., Wang, P., Goel, L., Billinton, R., Karki, R.: ‘Reliability Table 7 Electrical characteristics
assessment of restructured power systems using reliability network
equivalent and pseudo-sequential simulation techniques’, Electr. Operating condition Standard testing condition NOCT
Power Syst. Res., 2007, 77, pp. 1665–1671 (STC)
16 Karki, R., Billinton, R.: ‘Cost-effective wind energy utilization for
reliable power supply’, IEEE Trans. Energy Convers., 2004, 19, open-circuit voltage, 44.8 V 40.8 V
pp. 435–440 Voc
17 Billinton, R., Li, W.: ‘Reliability assessment of electric power systems short-circuit current, Isc 8.33 A 6.74 A
using monte carlo methods’ (Plenum Press, New York, 1994) maximum power, Pmax 280 W 204
18 Mello, J.C.O., Leite da Silva, A.M., Pereira, M.V.F.: ‘Efficient W
loss-of-load cost evaluation by combined pseudo-sequential and state
transition simulation’, IEE Proc. Gener. Transm. Distrib., 1997, 144,
pp. 147–154 STC: irradiance 1000 W/m2, module temperature 25°C, AM = 1.5
19 Leite da Silva, A.M., Da Fonseca Manso, L.A., De Oliveira Mello, J.C., NOCT: irradiance 800 W/m2, module temperature 20°C, AM = 1.5
Billinton, R.: ‘Pseudo-chronological simulation for composite reliability
analysis with time varying loads’, IEEE Trans. Power Syst., 2000, 15,
pp. 73–80 sequence is to simulate a series of states following the
20 Mello, J.C.O., Pereira, M.V.F., Leite da Silva, A.M.: ‘Evaluation of
reliability worth in composite systems based on pseudo-sequential
examined state until a success state is found.
Monte Carlo simulation’, IEEE Trans. Power Syst., 1994, 9, The frequency of departure from state s to state j is
pp. 1318–1326
21 Subcommittee, I.P.M.: ‘IEEE reliability test system’, IEEE Trans. fsj = ps lsj (32)
Power Appar. Syst., 1979, PAS-98, pp. 2047–2054
22 Masters, G.M.: ‘Renewable and efficient electric power systems’ (John
Wiley & Sons, Hoboken, NJ, 2004) where ps is the probability of state s, λsj is the transit rate from
23 Aguiar, R., Collares-Pereira, M.: ‘TAG: a time-dependent, s to j.
autoregressive, gaussian model for generating synthetic hourly For state s which can transit into number of following states
radiation’, Solar Energy, 1992, 49, pp. 167–174 (NFS) number of following states, the frequency of departure
24 Goh, T.N., Tan, K.J.: ‘Stochastic modeling and forecasting of solar
radiation data’, Solar Energy, 1977, 19, pp. 755–757 from state s is
25 González-Longatt, F.M.: ‘Model of photovoltaic module in matlab’, II
CIBELEC, 2005, 2005, pp. 1–5 
NFS
26 Singh, C., Lago-Gonzalez, A.: ‘Reliability modeling of generation fsdep = ps lsj (33)
systems including unconventional energy sources’, IEEE Trans. j=1
Power Appar. Syst., 1985, PAS-104, pp. 1049–1056
27 Stember, L., Huss, W., Bridgman, M.: ‘A Methodology for photovoltaic
system reliability & economic analysis’, IEEE Trans. Reliab., 2009, 31, The probability of system transiting from state s to j is then
pp. 296–303 determined as
28 C. University, Elia, E. TSO, H. Transmission, REE, and RTE. (2011).
D22; D331; D332; D32 Current Designs and Expected Evolutions of 
NFS
Day-ahead, Intra-day and Balancing Market/ Mechanisms in Europe. psj = fsj /fsdep = lsj / lsj (34)
Available at http://www.optimate-platform.eu/downloads/
j=1
29 Zimmerman, R.D., Murillo-Sanchez, C.E.: ‘Matpower 4.1 User’s
Manual’
30 Ding, Y., Wang, P.: ‘Reliability and price risk assessment of a By sampling based on a uniform distribution, the following
restructured power system with hybrid market structure’, IEEE Trans. state is determined according to (34).
Power Syst., 2006, 21, (1), pp. 108–116
The difference of previous state and present state is the
same as that for the forward state transition. The backward
8 Appendix 1: parameters of PV panel sequence is identified by simulating a series of previous
states of the examined state until a success state is found.
The probability distribution of the previous state is
The PV panel produced by Suntech Power Holdings Co., Ltd determined by using the same method. The frequency that
with model number of STP280-VRM-1 was used. The the system state s is transmitted from state i is
parameters used are listed below. Cell area of the panel is
about 1.65 m2. (See Tables 6 and 7). fis = pi lis (35)

For number of before/previous states (NBS) possible previous

9 Appendix 2: forward/backward sequence
states, the frequency that they arrive at state s is
simulation

NBS
The forward and backward sequences are sampled according fsarr = pi lis (36)
to the probability distributions of the system transition states. i=1
The difference between the present state and the following
state of the system is the operating state change of one The probability that system has transited from state i to s is
component which is either repaired or failed. The forward

NBS
pis = fis /fsarr = pi lis / pi lis (37)
i=1
Table 6 Temperature characteristics
Nominal operating cell temperature (NOCT) 45° ± 2°C Assume that states i and s are differentiated by component k
which is transformed from operating state to outage state. If
temperature coefficient of Pmax −0.44%/°C k is represented by a two-state Markov model, then
temperature coefficient of Voc −0.33%/°C
temperature coefficient of Isc 0.055%/°C      
pi /ps = mk lk + mk / lk lk + mk = mk /lk

IET Gener. Transm. Distrib., 2014, Vol. 8, Iss. 3, pp. 421–430 429
doi: 10.1049/iet-gtd.2013.0340 & The Institution of Engineering and Technology 2014
www.ietdl.org
and also λis = λk. If all the components are represented using whose state durations follow exponential distribution, the
two-state Markov models, we have ( pi/ps)λis = μk = λsi. expected duration of state j is
Dividing (37) by ps, we have
  
E Dj = 8760/ lk (39)

NBS
k
pis = lsi / lsi (38)
i=1
where λk is the transition rate between the examined state and
Note that (38) is similar with (34). any state it can transit to.
The duration of the outage state is determined by the The expected duration of the sequence I is then
combination of forward and backward sequences. The total   j
duration D of the sequence is the sum of all duration D j of E(D) = E D (40)
each failure state j. For statistically independent components j[I

430 IET Gener. Transm. Distrib., 2014, Vol. 8, Iss. 3, pp. 421–430
& The Institution of Engineering and Technology 2014 doi: 10.1049/iet-gtd.2013.0340