Reliability analysis of electric distribution grid by Monte Carlo method
A.T. Sikora, S. Rusek, V. Král & R. Goňo
Department of Electric Power Engineering, VSB, Technical University of Ostrava, Ostrava, Czech Republic
ABSTRACT: The paper deals with the description and application of a new program for the calculation of the
reliability of the electric power networks (of all voltage levels) by using the Monte Carlo method. The program
was elaborated at the Department of Electric Power Engineering of VŠB - Technical University of Ostrava. It
enables the network to be modeled by using typical reliability data of individual elements, to define reserves by
means of reserve branches and reserve areas, as well as to visualize the network graphically including the set
and the calculated values. Both possibilities and properties of the program are illustrated on the example of a
high voltage electricity distribution network. In this example were calculated variants of network configuration
and their influence on resulting reliability of supply. Next part of the paper describes convergence of results
depending on number of Monte Carlo iterations.
1 INTRODUCTION – Breakdown service logistics is getting better – the
failure duration is shorter.
Simulation calculations of electric power systems are – The range of live servicing is getting bigger –
well known and used all around the world for sev- minimizes the maintenance downtime duration.
eral years. What have been changed during this period – The reliability centered maintenance (RCM) is
and what more can be written about this topic? In implemented – maintenance downtime duration is
this paper we are trying to briefly resume changes optimized and in many cases is no longer constant
in simulation calculations in past decade considering value for one type of component.
reliability calculations of electric networks. Further a
All that mentioned above exert influence on simu-
new computer program for electric grid’s reliability
lation process. The RCM implementation and Remote
calculation is introduced. That software is developed
controlled components installation forces modifica-
at the Department of electric power engineering at
tion of whole simulation algorithm.
VSB-TU of Ostrava. At the end, last but not least, the
results properties of simulation reliability calculation 3 PROGRAM FOR RELIABILITY
are described. CALCULATION BY SIMULATION MEANS
2 WHAT IS NEW IN SIMULATION
CALCULATIONS? 3.1 Calculation procedure
The calculation is executed by using the sequential
In the very first place the computer equipment, on Monte Carlo simulation technique. For both branches
which simulation runs, have been changed. In the past and nodes are used these reliability parameters:
ten years almost all computer technical parameters
are about thousand times bigger: processor frequency, failure rate λF (1/year),
memory, hard drive space etc. For the simulating calcu- maintenance rate λM (1/year),
lation it means that in the same time period we are able failure duration τF (hour),
to simulate more complex grid, or run more iterations maintenance duration τM (hour).
and make the result more accurate. Second advantage Program uses also other parameters for grid model
is, that we need no longer to input data for simulation (e.g. node load, branch load capacity, peak demand
from command line, but set the network from graph- utilization hours…), but these parameters were not
ical environment, or by importing most values from employed in calculation described in this paper.
some other program, e.g. for programs for steady state The outage rate λgs calculated as simple sum of
calculation. failure rate and outage rate (as in series reliability sys-
Nevertheless the networks are changing as well, tem. The outage duration τ is calculated as weighted
especially these in former “East Block” countries: average of both failure and maintenance duration
– Modern and more reliable components are according to following formula:
installed – failure rate decreases.
– Remote-controlled components are installed – the
outage duration is shorter.
1951
© 2010 Taylor & Francis Group, London, UK
At the beginning of simulation, it is supposed that
all components of the network are in operation. Then
it is necessary to generate a random variable, which
represents a failure of a component, for each com-
ponent. Program mainly uses exponential distribution
function, f (t) = gλe−λt , the sampled state duration
(time-to-failure) is then T = −ln(U )/λ, where U
is uniformly distributed random number (0,1>. The
outage time is considered constant τ.
The smallest one of variables (time-to-failure T or
outage time τ) is chosen (that occurs first); a relevant
change in the model is made and a new condition of
the system is evaluated. For the component affected
by the change a new value is generated – this time
putting into operation – and is queued into waiting
events. The system time is transferred to a new point –
the instant of change. From the queue of waiting events Figure 1. Reliability diagram of tested network.
the first one is chosen and the whole procedure repeats
until the required number of tests is performed. If the Table 1. Reliability parameters of branches.
adequate number of tests is carried out (the variation
of results is less than set-up value, or iteration count λF τF λM τM
Branch Branch type
achieved desired value), it is necessary to execute final name
operations. (year−1 ) (h) (year−1 ) (h)
Detailed description of simulation process can be
V1 Transformer 0,02 30 1 120
found in Sikora (2007). Similar technique is described V2 Line + backup 2,62 4,23 1 200
also in Billington & Wangdee (2006). V4 Line 1,58 4,72 1 120
V5 Line 1,58 4,72 1 120
3.2 Node reliability evaluation V6 Line 1,05 4,41 0,333 80
V7 Line 1,05 4,41 0,333 80
The evaluation of the reliability of supplies to the nodes V8 Line 0,53 5,32 0,333 16
is, as a matter of fact, the main reason why the calcu- V9 Transformer 0,02 30 1 120
lation is performed. However, it is not as simple as in V10 Line + backup 2,1 4,41 1 160
the case of branches. V12 Line 3,14 4,11 1 240
To be able to decide whether the node is in normal
operation, or in outage, one must evaluate the condi-
tion of the whole network at each step of simulation. hour. The numbers at other switches showing time of
If the node has the demand Pi and the sum of load car- switching operations (in hours).
rying capacities of branches leading to this node "Pij
is greater than this demand, all is right. In the contrary 4.2 Calculation variations
case the node is considered disconnected.
Step-by-step – for each trial a 1 mil. iterations have
4 RESULTS TESTING OF THE SIMULATION been done, results saved, then continue for next 1
PROGRAM million and so on up to 5 mil. iterations.
Standalone calculation – each trial was calculated
This chapter aims to describe “random” reliability form start, each one for 1 million iterations.
calculation results of the program ComplexSPOLEH. Calculation with Stand-by area and with No backup.
Mainly the functionality of standby areas is verified. 4.3 Step-by-step calculation
Also the results variation is illustrated according to
The results of this test are on Fig. 2. At the first two
calculation step-by-step or 5 standalone simulations.
nodes (U2 & U3) is noticeable rising trend of reliabil-
At the end the convergence of results is shown.
ity, while reliability value of the rest of nodes varies. At
this range, however, trend is purely random. The reli-
4.1 Tested network
ability value varies only on the second place after the
Tested network in Fig. 1 contains all components, last “9”, what means error (or in better words variation)
with which the ComplexSPOLEH program is able to in units of percents. Therefore it can be said, that trends
calculate. There are: simple branches (transformers, are negligible and inconclusive.
single lines), both hot (parallel lines) and cold (par- In the finer scale, as can be seen in Fig. 3, is “trend”
allel line with switching operation) reserves and of reliability change at the U2 and U3 nodes much
standby area. more expressive. Nevertheless the, finer scale reveals
(Data acquired based on Gono et al. 2009). also reliability change only on the second place behind
The Stand-by area is after failure of main area con- last “9”, and that only within one point. Variation of
nected to node U 4 with switching time TmanO = 0, 25 unreliability result is therefore maximum 1 %.
1952
© 2010 Taylor & Francis Group, London, UK
Figure 2. Reliability for step-by-step calculation with Stand-by area for trials 1 to 5.
Figure 3. Detail of nodes U2 & U3.
Figure 4. Reliability for standalone calculation with stand-by area for trials 1 to 5.
step-by-step calculation there is no trend aiming on
possibility of convergence of results to any value.
4.5 Standby areas
The aim of this test is to confirm positive influence
of standby area on network’s reliability. The results
comparison of network with and without standby area
is shown on Fig. 5.
As shows Fig. 5, the presence of standby area has
Figure 5. Node reliability with Stand-by area and with No positive influence on overall network reliability (right
backup. part of the graph).
4.6 Statistical evaluation of simulation results
4.4 Standalone calculation
A value of unreliability was selected for statistical eval-
A more significant variation of results is at the uation, because the reliability, as a value nearby one,
results of standalone calculation trials, as can be seen would not has a corresponding efficiency in relative
on Fig. 4. (percentage) values. Only a little corresponding effi-
It means that 1 million iterations are not ciency has also the determination of standard deviation
enough, even for such simple network. Contrary to in absolute values (see Fig. 6).
1953
© 2010 Taylor & Francis Group, London, UK
with resource area (the third and the fourth group
of values). The reason is the fact, that nodes without
standby area has higher unreliability, thereby standard
deviation in absolute numbers is much higher.
On the Fig. 7 there is correct evaluation with rela-
tive values (standard deviation to average, by 100%):
the biggest error (standard deviation in %) has the
calculation without continuation with resource area
(the third group of values), then the calculation with
continuation with resource area (the fourth group of
Figure 6. Standard deviation of unreliability. values). Results of grid with No backup have much
lower standard deviation – all values of nodes are less
one percent.
4.7 Convergence test
A convergence test has been preceded on that same net-
work. Results of each trial have been saved every 10,
20, 50, 100, 200, 500, 1000…up to 109 iterations and
the values of unreliability of Demand node have been
compared. As reference the result of test with 109 iter-
ations has been taken and for each result the accuracy
Figure 7. Percentual standard deviation of unreliability.
calculated. The outcome is depicted in Fig. 8 & 9.
The logarithmic graph in Fig. 9 shows accuracy
results for each test 1-5 and the average. The accuracy
rises hyperbolically with growing number of itera-
tions (in logarithmic graph it looks like the error sinks
approximately linearly with order of iteration number.
So if we need reliability with at most 1 % error, then
105 iterations will be enough.
5 CONCLUSION
Figure 8. Unreliability percentage error – semi-logarithmic Simulated calculations are still acceptable instrument
axes. for calculations of reliability complicated electrical
nets, which do not have analytical solution, or only
with big problems. But a result of simulation still
depends on man, who create simulative models and
who has to understand that model. Another question
is to correct interpretation of results, because you can
have also with good results wrong conclusion.
ACKNOWLEDGEMENT
This work was supported by the Ministry of Education,
Youth and Sports of the Czech Republic (No. CEZ
MSM6198910007).
Figure 9. Unreliability percentage error – logarithmic axes.
Due to expectance, at Step-by-step calculation has REFERENCES
a smaller standard deviation for all nodes (the second
and the fourth group of values on the Fig. 6), than a Gono R., Kratky M., Rusek S. 2009. Analysis of Failures in
Standalone calculation without continuation (the first Electrical Distribution System. Inproceedings of Electric
and the third group of values). The reason is because Power Engineering. 2009. Ostrava, Czech Republic: pages
of more iteration are done at Step-by-step calculation 379 – 383
Sikora, Tadeusz 2007. Matematické metody výpoètu
and thereby also more accurate results. spolehlivosti a spolehlivostních ukazatelù (Mathemat-
There is misguided information follow on the Fig. 6, ical Methods of Reliability and Reliability Indicators
even if the network without standby area (the first and Calculation). Ph.D. Thesis, VSB-TU Ostrava
the second group of values) was simpler and after cal- Billinton R., Wangdee W. 2006. Delivery Point Reliability
culation on the number of component units there was Indices of a Bulk Electric System Using Sequential Monte
more iteration and the result should be more accurately, Carlo Simulation. IEEE Transactions on Power Delivery,
it has worst standard deviation than the calculation 21(1), Pages 345 – 352
1954
© 2010 Taylor & Francis Group, London, UK
