IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 26, NO.

2, MAY 2011 905

A Novel Hybrid Network Architecture to Increase

DG Insertion in Electrical Distribution Systems
Marie-Cécile Alvarez-Hérault, Student Member, IEEE, Damien Picault, Raphael Caire, Member, IEEE,
Bertrand Raison, Member, IEEE, Nouredine HadjSaid, Senior Member, IEEE, and Wojciech Bienia

Abstract—Distribution networks will experience a deep mu- to load. Nevertheless, concerning consumption and production
tation concerning their planning and operation rules due to the conditions, DGs have been reported to have considerable im-
expected increase of distributed generation (DG) interconnection pacts on the distribution network [3]. First, DGs can modify
to the grid. Indeed, the opening of the electricity market or the
growing global concern for environmental issues will lead to a the electrical values such as voltages, currents, and power flow.
massive development of DGs. Yet, a too large amount of DGs could The voltage profile in the distribution network depends on both
raise technical problems on distribution networks which have the injected active and reactive power by DGs and the loads.
not been planned to operate with bi-directional power flow. The It is presently well known that the interconnection of DGs can
existing solutions to solve marginal DG connections could be no lead to the violation of voltage limits, to the dysfunction, and
longer relevant. The distribution network definitely has to evolve
towards a smarter and more flexible network. Two possible ways even the deterioration of the network components. In addition,
to reach this goal are through new architectures and developing power flows that were unidirectional, coming from the transmis-
intelligent systems. This paper focuses on new architectures and sion network to the end users, can be modified by the injected
operating modes. The traditional radial distribution network power of DGs. Consequently, in some circumstances, the max-
could accept more DGs by introducing appropriately specific imum allowed current of a conductor can be exceeded. If the
loops. A new hybrid structure enabling the coexistence of the
radial and meshed operation is proposed. It is equipped with au- production of DGs is greater than the global consumption, the
tonomous circuit-breakers and automated switches that improve distribution network could export power. The short-circuit cur-
its reliability. A heuristic algorithm is also proposed to build this rents can be modified and lead to undesired behaviors of the
new architecture while ensuring the equality of consumers with protection scheme (e.g., protective relay blindness or inoppor-
respect to the continuity of service and while minimizing the global tune trips). Furthermore, DGs supply the short-circuit current
cost.
that may exceed the operating limits of the network elements
Index Terms—Distributed generation, distribution network, op- [4]. The quality of the voltage can also be reduced [3].
timization, planning, Traveling Salesman Problem.
In many countries, in order to face these problems, the
reinforcement of the network or the dedicated feeders is used.
I. INTRODUCTION The first solution consists in detecting the part of the network
where constraints are violated. In case of current (or voltage)
HE distributed generators (DGs) are small production violation, the gauge mutation of conductors can annihilate the
T units based either on renewable energy sources (such as
wind and solar photovoltaic) or conventional energy (such as
problem. The second solution consists in connecting the DG
with a dedicated feeder directly to the HV/MV substation. By
small gas engines or diesel generators) that are connected to the doing so, the power flow becomes unidirectional in normal
distribution network. Their installed capacity reaches 10 MW operation mode.
in the USA and 12 MW in France [1], [2]. Although connecting marginal quantities of DGs in the dis-
In the near future, the growing concern for environmental is- tribution network is currently well managed using “business as
sues as well as for the security of supply is expected to lead to usual” techniques (reinforcement and dedicated feeder), the sys-
the development of local renewable DGs. These power sources tematic use of dedicated feeders could become a very expensive
will be connected to the distribution network which had been solution to manage while considering a significant development
designed to see only unidirectional power flows, from upstream of DGs.
New solutions must, thus, be found to assist the introduc-
Manuscript received March 19, 2010; revised March 22, 2010 and July 30,
tion of a large amount of DG in the distribution network. Some
2010; accepted August 16, 2010. Date of publication September 23, 2010; date studies proposed to change the operation mode of the distribu-
of current version April 22, 2011. This work was supported by the Carnot Insti- tion network, inspired by the transmission network, which uses
tute ’Energies du Futur’, Grenoble, France. Paper no. TPWRS-00211-2010.
M.-C. Alvarez-Hérault, D. Picault, R. Caire, B. Raison, and N. HadjSaid are
a meshed operation. This solution has been considered as in-
with the Grenoble Electrical Engineering Laboratory (G2Elab), Saint Martin applicable for many years due to the parallel operation of trans-
d’Hères, France (e-mail: alvarez@g2elab.grenoble-inp.fr; picault@g2elab. mission network and distribution network. Such a solution could
grenoble-inp.fr; caire@g2elab.grenoble-inp.fr; raison@g2elab.grenoble-inp.fr;
hadjsaid@g2elab.grenoble-inp.fr).
prove to be dangerous without implementing an appropriate pro-
W. Bienia is with the Laboratory of Grenoble for Sciences of Conception, tection scheme since loops currents are created. Recent studies
Optimisation and Production (G-SCOP), Grenoble, France (e-mail: Wojciech. have proposed to use current limiters or D-FACTS [5] that limit
Bienia@imag.fr). loop and short-circuit currents [6]. Further studies report that
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org. meshing could be a competitive way to increase the connection
Digital Object Identifier 10.1109/TPWRS.2010.2071887 of DGs in the distribution network [7]–[13].
0885-8950/$26.00 © 2010 IEEE
906 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 26, NO. 2, MAY 2011

Fig. 2. Hybrid structure.

Fig. 1. Secured feeder.

the loop [14], which are also radially operated. The redundant
paths increase the service quality but the operation of such struc-
This paper proposes a new architecture called hybrid structure tures can be very complex, and can also be very expensive.
which can increase DG connection to the network. This archi- In the future, the parameter “distributed generation” should
tecture is an evolution of an existing architecture which is ra- be integrated in the objectives of the long-term planning. Hy-
dially operated which introduces partially meshed sub-feeders. bridization between the secured feeder and the permanent loops
Beginning from the location of HV/MV and MV/LV substa- could be a novel way to increase DG connection, ensuring the
tions, an algorithm that builds the hybrid structure ensuring the service quality and minimizing the global cost.
equality of consumers regarding the energy supplied while min-
imizing the total length of conductors is proposed. This algo- B. Hybrid Architecture
rithm has been applied on an existing urban network and the
The hybrid structure is illustrated on Fig. 2. As in the se-
capacity of the hybrid structure to accommodate DGs is high-
cured feeder, consumers are supplied by radially operated main
lighted.
feeders using normally open switches. But the main feeders are
II. NEW ARCHITECTURE TO INCREASE DGS not directly connected to the consumers. They are connected
INSERTION: THE HYBRID STRUCTURE with other consumers or producers on secondary circuit (sub-
feeders) which are looped.
A. Current French Network Architecture In case of a fault on the main feeder and after the trip of a
The main challenge of distribution long-term planning was to circuit breaker within the HV/MV substation, the network is
find the best architecture ensuring a good service quality while rearranged by changing the state of the normally opened and
minimizing the global cost. This challenge has lead to the se- closed switches. In case of a fault inside a loop, a “distributed”
cured feeder structure [14] which is the most encountered ar- circuit-breaker disconnects only the loop so that the main feeder
chitecture in France and many countries, as depicted in Fig. 1. and the other loops are not affected by the fault. Each MV/LV
Each consumer is supplied by several main feeders that link two substations in the loop can be equipped with automated switches
HV/MV substations. The radial operation is ensured by using that can quickly locate and isolate the fault. The loop can be,
normally open switches in every main feeder. A main feeder then, reconnected to the main feeder in order to reenergize the
consists of two radial feeders usually limited to 6 MW. healthy parts [14].
The desired service quality leads to a given number of main The procedure to reenergize consumers in case of a fault is
feeders and the choice of the consumers connected to them. The faster than that of the secured feeder. Therefore, the SAIDI
usual service quality notions include: and the EENS decrease. The partially meshed operation will in-
— System Average Interruption Duration Index (SAIDI) ex- crease DG connection. Finally, supply and consumption balance
pressed in minutes per year; can be ensured by defining PL equivalent areas. As in the se-
— System Average Interruption Frequency Index (SAIFI) ex- cured feeder, each main feeder and each main feeder have equiv-
pressed per year; alent PL products and each loop has an equivalent PL.
— expected energy not supplied (EENS) expressed in kWh
per year; C. Definition of the Evaluation Indexes
— equality of the consumers regarding the EENS. This index The building of the hybrid structure consists in linking con-
is ensured by creating areas with equivalent product PL. sumers and producers respecting the balancing of PL product.
For a given area, the product PL is the product of the total power This structure is expected to increase the DG connection, to en-
demand (in MVA) and the total length of conductors connecting sure the quality of service (SAIDI and SAIFI) and to be com-
consumers to HV/MV substations (in km). If this index is bal- petitive. In this subsection, the different indexes (balancing the
anced, areas having a small consumption will be supplied with PL product index, global cost, SAIDI, SAIFI, and maximal DG
longer conductors than areas having a higher consumption. The insertion rate) are mathematically expressed.
probability that a fault occurs on the network is proportional to 1) Balancing the PL Product: The PL balance is expressed
the length of conductors. Thus, the statistical power cut during by using the standard deviation E of the PL product regarding
a fault is minimized if the PL is balanced. The service quality is the mean of the PL product. Formula (1) gives the expression of
improved. Consequently, the EENS of two areas with the same criterion E:
PL will be statistically the same. In the secured feeder, each
main feeder has the same product PL.
In areas where the expected service quality is high, the se-
cured feeder can evolve into two main structures: the grid and (1)
ALVAREZ-HÉRAULT et al.: A NOVEL HYBRID NETWORK ARCHITECTURE TO INCREASE DG INSERTION IN ELECTRICAL DISTRIBUTION SYSTEMS 907

where of the fault: on the main feeder or the loop. Since the operation
of these two cases is different, the SAIDI and SAIFI calculation
standard deviation of PL regarding the mean (%);
will be different.
number of areas; If a fault occurs on a main feeder, the breaker located at the
product PL of area i (MVA.km); top of the feeder will trip and each customer depending on the
feeder is disconnected. In a few minutes, remotely controlled
switches isolate the fault and reenergize all the customers. The
SAIDI and SAIFI due to a fault on the feeder i are explained
(2) with the formula (6):
Minimizing E ensures the balancing of PL and then the equality
of consumers regarding the service continuity.
(6)
2) Various Costs of the Network: The cost of the network is
calculated by using the principle of actualization. It thus takes
where
into account the fluctuation of money on the life duration of
the network through the actualization rate . is a fixed value
failure rate for the feeder per year per km;
expressing the fact that N euros at the year N will be equiva-
lent to euros at the year . Then, if a utility length of the feeder i in km;
decides to invest D(t) euros at the year t, this investment costs duration to isolate the fault with the remotely
euros today. The actualization cost has been fixed controlled switches.
at 8% [15]. The different costs of the networks need to be actual-
ized to ensure correctness. The formula (3) gives the actualized If a fault occurs inside a loop, the breaker disconnects the
cost of an annual cost on N years: loop from its main feeder. Therefore, the other customers are not
affected by the fault. Then automated switches enable the fault
to be isolated. The total SAIDI and SAIFI due to a fault inside
(3) every loop of the feeder i are presented in (7) at the bottom of
the next page, where
The actualized power losses cost represents the cost of the
number of loops of the feeder ;
power losses during N years, where N is planning period:
duration of the automatisms of the MV/LV
normally closed switches of substations of
(4) the loops (20 s). There are two normally
closed switches per MV/LV substations.
where Each operation of a switch (opening or
closing) takes 10 s;
actualized power losses cost in euro; length of the loop j of the feeder ;
investment at the year 0 in euro; number of customers of loop of the feeder ;
cost of one kW of losses at peak time in euro; number of MV/LV substations of loop of
power losses at peak time at the year n in kW; the feeder ;
actualization rate in . number of customers of feeder .

The actualized failure cost includes the EENS cost and Finally, the average SAIDI and SAIFI of a feeder is presented
the outages cost. The EENS cost was chosen as high as 9.5 in formula (8):
euro/kWh and the outages cost at 0.8 euro/kW [16]. Conse-
quently, the actualized failure cost is given by the formula (5):

(5)

where
expected energy not supplied;
(8)
total number of outages;
maximal consumption; where is the number of feeders of the network.
number of hours of maximal consumption. 4) Maximal DG Insertion Rate: The maximum DG insertion
rate of the hybrid structure is defined with formula (9):
3) Reliability Criteria: To evaluate the reliability of
the hybrid structure, the SAIDI and SAIFI (defined in part
Section II-A) are relevant and international indicators. The
(9)
reliability indicators are computed by considering the location
908 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 26, NO. 2, MAY 2011

where
maximal insertion rate of DGs;
maximal power output of DG;
total maximal consumption of the network.
The maximal insertion rate of DGs is the maximum power of
DG that can be connected to the distribution network without
violating technical constraints (such as voltage and current
values). A Monte Carlo algorithm has been used to estimate
this rate without making any hypothesis on the type, the size,
and the location of DGs [17]. To satisfy these hypotheses, each
simulation consists of checking the network constraints in the
worse state (minimum consumption and maximal production).
This situation increases bidirectional power flows in the net-
work and thus increases the probability of technical constraints
violation.
The estimation of the maximal DG insertion rate for a given
network is detailed in Fig. 3. This algorithm will be illustrated
with the following example:
— total consumption: 100 MW;
— number of loads: 1000.
Using the previous example for , it represents an intro-
duction of 10 MW of DG in the network. One possible scenario
could be 5 DGs of 2 MW, another could be 10 DGs of 1 MW,
and so on. One thousand scenarios are carried out and lead to
950 successes for example. Then .
The probability that the technical constraints are respected in- Fig. 3. Monte Carlo algorithm.
troducing 10 MW of DGs in the network is 0.95.
The maximal DGs insertion rate is evaluated using the max-
imal rate whose probability of success is higher than a given layout is conducted (sizing and placing the different conductors
probability. For instance, for a given risk of 3%, this probability to reach a hybrid structure).
is 0.97. This risk means that the hypothesis of minimal con-
sumption and maximal production is unlikely. This method is A. Traveling Salesman Problem (TSP) and the Vehicle Routing
well suited because it allows to evaluate if an architecture is fa- Problem (VRP)
vorable to DGs insertion without making any hypothesis on the The problem of building the hybrid structure consists in con-
composition of DGs (solar, wind, etc.) and the location. Indeed, necting a group of consumers (MV/LV substations) with pro-
these two last parameters are hardly predictable. ducers (MV/LV substations) using loops. This problem is sim-
ilar to the Traveling Salesman Problem (TSP) and the vehicle
III. MODELING THE HYBRID STRUCTURE CONSTRUCTION routing problem (VRP) [18]. The TSP is a classical combina-
The planning problem presented in this paper considers torial optimization problem. For a given number of cities, the
only the consumers (MV/LV substations) and the HV/MV TSP consists in finding the cheapest tour that visits every city
substations locations. Then, the complete rebuilding of the grid exactly once. The cost of each edge is a value associated to this

(7)
ALVAREZ-HÉRAULT et al.: A NOVEL HYBRID NETWORK ARCHITECTURE TO INCREASE DG INSERTION IN ELECTRICAL DISTRIBUTION SYSTEMS 909

lutions space in order to identify the best sub-location to search.

Those methods allow the deterioration of the solution and even
solutions very far from the optimum in order to prevent the al-
gorithm from converging into a local minimum. Several meta-
heuristics can be applied to solve TSP [18]: simulated annealing
[25], deterministic annealing [26], tabu search [27], genetic al-
gorithms [28], ant algorithms [29], and neural networks [30].
Fig. 4. Contradiction between the criteria L and PL. The solutions are more accurate than those of the heuristics
methods, but the computation time is longer. Reference [18]
tests and sorts the different algorithms on different examples.
path (for instance, the length of the edge). A notion of limited In addition, the parameters of those methods need thin adjust-
capacity can be added to each edge. Also, several tours are nec- ments that depend on the studied problem.
essary to visit each city. This is the Vehicle Routing Problem 2) TSP: All the above heuristic and metaheuristic methods
(VRP). A classic example of the VRP is the dustman problem can be applied and also greedy algorithms to solve the TSP. So-
whose packer body is limited [18]. lutions and computation time differ from the chosen algorithm.
Both in the TSP and VRP, the objective function (the total A well-known algorithm is the Christofides one [19]. It is faster
cost of the tour) is linear. In the hybrid structure problem, the than many existing methods, and it is the only one that guaran-
standard deviation E of product PL regarding the mean and the tees that the solution is far from the optimal one from 50% [19].
total length L of the conductors are to be minimized. But, the Indeed the other algorithms give a solution without any accu-
criteria E is nonlinear and E and L are contradictory criteria. racy on its reliability. Then, the given solution can be very far
This phenomenon can be illustrated with a simple example of from the optimal one.
a network with two HV/MV substations and 14 MV/LV sub- 3) Hybrid Structure Problem: The difference between our
stations. The load number 1 consumes 100 MVA and the other problem and the TSP and VRP is that the objective function is
loads 5.58 MVA. The objective is to build two main feeders. In multicriteria and nonlinear. Indeed, the PL product (nonlinear
case 1, the aim is to minimize the total length of conductors and criterion) has to be balanced while minimizing the total length of
in case 2, the aim is to balance PL products. Fig. 4 shows that conductors. Constructive heuristics or two-phase heuristics pre-
balancing the PL product leads to a solution with a total length viously described are not suitable because it is impossible to at-
of conductors approximately 6 km higher. tribute a cost to an edge that takes into account the nonlinear cri-
The hybrid structure problem seems to be more complex than terion PL. Metaheuristic algorithms would need to be adjusted
the TSP or VRP. The recognition version of the TSP and the for each case because of the distribution diversity of the loads.
VRP are NP-complete problems [19]. In other words, there is Nevertheless, the improvement method seems to be appropri-
not any algorithm that can find the optimal solution in a poly- ated. First, in stage 1, the loops will be built with Christofides
nomial time. Only good local sub-optimums can be obtained version of the TSP [19]. The improvement algorithm will bal-
using smart heuristics. The building of the hybrid structure is a ance the PL product of those loops. Then, in stage 2, the loops
derivative of the TSP and the VRP so it is also NP-complete. will be connected with a nonlooped version of the TSP [31]. The
An overview of existing methods to solve the VRP and TSP is same improvement algorithm will balance the PL product of the
necessary to find an appropriate heuristic algorithm to build the main feeders.
hybrid structure.
IV. PROPOSED ALGORITHM TO BUILD THE HYBRID STRUCTURE
B. Comparison of the Existing Algorithms
1) VRP: Two main classes of heuristics attempt to solve the A. Objectives
VRP: classical heuristics (mainly developed between 1960 and Considering only the data on HV/MV and MV/LV substa-
1990) and metaheuristics developed during the last two decades. tion locations, the objective is to automatically generate the con-
More information on those methods can be found in [18]. Clas- ductor gauges and paths to supply the consumers (MV/LV sub-
sical heuristics are local search methods that do not explore all stations) following the hybrid structure. The location of the con-
the space solutions. They enable to get acceptable solutions in ductors will be created by minimizing the criterion E detailed in
a reasonable time and can be easily adapted to many problems (1). Minimization of the total length of conductors will be en-
of daily life. Three categories of heuristics are used to solve the sured by the TSP. This will lead to minimize the investment cost
VRP [18]. First, the constructive methods (for instance Clark of the network. Considering the reliability, the number of loops
and Wright [20], Christofides, Mingozzi, and Toth [21]) built a will directly impact the failure cost. Then the final number of
feasible solution while keeping an eye on the cost of the solu- loops of the structure will be a compromise between the invest-
tion. Then, two-phase methods (for example, the sweep algo- ment costs and the quality of service desired.
rithm [22], Fisher and Jaikumar [23]) divide the problem into The technical constraints (current and voltage) will be sat-
two phases. First, each vertex is allocated to a vehicle. Then, isfied by choosing the gauges of the cables following normal-
several routes are built. The reverse method is also possible [24]. ized values. Choosing high gauges will decrease power losses.
Finally, improvement methods try to improve any possible so- So, even if the initial investment is higher, the global actualized
lution making exchanges of vertices or edges (cross, exchange, cost of the network could be cheaper. A load flow is then used
or relocation [18]). Metaheuristics consist in exploring the so- to verify the technical constraints of the studied network.
910 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 26, NO. 2, MAY 2011

Fig. 5. Example of an inclined checkerboard.

B. Description of the Proposed Algorithm

The proposed algorithm used both in stages 1 and 2 consists
in building an initial solution and then, using an improvement
method, minimizing the criterion E. The only difference be-
tween stage 1 and stage 2 is the TSP algorithm used, as described
in the following subsections.
1) Initial Solution Generation: A uniform partitioning is
automatically generated following an inclined checkerboard.
Fig. 5 shows an inclined checkerboard on a network of 31
MV/LV substations. This partitioning has been chosen first for
its simplicity and also because no assumptions have to be made
about power density of loads. The inclination of the checker-
board is given by the line joining the two HV/MV substations.
2) Improvement Algorithm: The improvement algorithm
used to achieve local exchange between loops is summarized
in Fig. 6.
The parameters used in this diagram are as follows.
— is the Euclidian distance of the load
of the loop j to the selected loop .
— is a two-column matrix. For each nonselected loop
j, the first column gives the minimum distance between the
given (or receiving) loop j with the receiving (or giving) Fig. 6. Improvement algorithm.
loop .
The second column gives the corresponding load that can
be exchanged between the loops. Step 1) Selection of the loop: The selected loop has the
— is a vector compound with one load per loop j highest absolute deviation. If this deviation is neg-
selected in the second column of each matrix with ative, then the loop should be expanded to include
the first biased wheel. more loads. If it is positive, some loads should be
— is the final load selected in the matrix with the moved to other loops. The selected loop is named
second biased wheel. LS.
Considering the initial solution, the criterion E and the devia- Step 2) Biased wheel: If the load exchanged was only the
tion of the product PL of each loop with respect to the mean are nearest to the selected loop, then the algorithm could
computed. This deviation (EC) is defined with the formula (10): fall in a local minima. No other exchanges could
be possible. The algorithm will be stopped, but the
final solution would not be satisfactory. Therefore,
(10) two biased wheels are set to choose which loop will
be selected and whose loads will be selected. Prob-
abilities have to be biased to favor loads that are the
where
nearest from .
— If has to be expanded, then for each neigh-
product PL of the loop ;
boring loop j and for each load of loop j,
mean of product PL of all the loops; the minimal distance between and
number of loops. the loop is computed. Then, the matrix
ALVAREZ-HÉRAULT et al.: A NOVEL HYBRID NETWORK ARCHITECTURE TO INCREASE DG INSERTION IN ELECTRICAL DISTRIBUTION SYSTEMS 911

the mean) are computed again. While EC is higher than a given

value fixed by the planning operator, loops keep on making ex-
changes.
3) Building of the Main Feeders: Once the loops are gener-
ated and their product PL is balanced, the main feeders have to
be built to supply the loops. For each loop, the loads that are
the closest to one of the two substations are selected. Each main
feeder will pass by the two substations and some of the loads se-
lected. The same heuristic method is applied to build the main
Fig. 7. Example of a biased wheel. feeders. The planning operator gives the number of main feeders
that are desired. Then an initial solution is generated, and then
the improvement method previously described is applied. This
is created. Its first column is , and in final part is extremely fast because there are very few loads to
the second one, the corresponding load is . supply (the number of loops) and because, for a selected feeder,
— If some loads of has to be transferred from a it has, at the worst case, only two neighboring feeders. The only
feeder to another one, then for each neighboring difference is the way to build the initial solution. Here, the goal
loop j and for each load M(i) of the loop , the is to link the two given HV/MV substations passing by some
minimal distance between M(i) and the other points. A simple algorithm based on an angular criterion is
loop is computed. Then, the matrix is better adapted to minimize the conductor length while avoiding
created. Its first column is , and in the crossing cases. The line joining the two HV/MV substations
second one, the corresponding load is M(i). The separates the space into the superior part and the inferior part.
matrix gives the parameter to build the For the superior part, starting for left HV/MV substation, the
biased wheel. first feeder is built stepwise taking the point that has the max-
To build the biased wheel, the probability that a load imal angle until the other HV/MV substation is reached. Then
is first selected is computed with the formula (11) at other feeders are created until all the points of the superior part
the bottom of the page. are supplied. For the inferior part, the method is the same. The
Results are illustrated with pie charts. A number can only difference is that the criterion to select loads is not the max-
be randomly chosen following a uniform law on the imum angle but the minimum one.
interval . The number chosen gives the load
of the loop that will be exchanged. For example in V. APPLICATION ON A REAL URBAN NETWORK
Fig. 7, if the number chosen is 0.23, then the selected The hybrid structure is built for several numbers of loops. The
load is the load number 5. There are as many biased criterion E representing the standard deviation of the product PL
wheels as the number of loops minus one. regarding the mean is a free parameter. The presented results
Step 3) Rebuilding of loops and noncrossing checking have been carried out for a fixed standard deviation of 10%. The
(planar graph). impact of the number of loops will be studied regarding several
The final selected load is exchanged. The Christofides algo- costs of the network, reliability criteria, and maximal DG inser-
rithm enables to rebuild the two updated loops minimizing the tion rate. The studied network, depicted in Fig. 8, is extracted
total length of conductors. A “not crossing test” is also applied from a real French urban network. It consists of two MV/LV
to check that the new loops do not generate crossing. The rule substations that supply 984 MV/LV substations and 85 239 MV
used is based on the fact that loops consist of several segments and LV customers consuming a peak power of 151.4 MW.
connecting two consumers. In the case of crossing, another load The hybrid structure is built with the algorithm explained in
is selected with the biased wheel. The criterion E (standard de- part IV. To compare the impact of the meshing level on the costs,
viation E of the product PL regarding the mean of the product reliability, and maximal DG insertion rate, the hybrid structure
PL) and EC (deviation of the product PL of each loop regarding has been built for several loops. The number of loops that can be

(11)
912 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 26, NO. 2, MAY 2011

Fig. 10. Reliability indexes regarding the number of loops.

Fig. 8. Studied network.

Fig. 11. Maximal DGs insertion rate regarding the number of loops.

TABLE I
Fig. 9. Evolution of the costs of the network regarding the number of loops. COMPARISON BETWEEN THE SECURED FEEDER AND THE HYBRID STRUCTURE

built is included in the interval . The lower limit is due

to maximal allowable current constraints. The possible conduc-
tors are chosen in a list of normalized conductor currently used
in France. The upper limit is due to the fact that a loop must
contain at least three nodes. The hybrid structure has been built
for the same number of main feeders in order to compare the
impact of the number of loops.

A. Different Costs of the Network

Fig. 9 gives the different costs previously described in part II
for a 40-year planning period. The conductors have been chosen
in [15]. The choice of conductors is made using the actualized The protection scheme and associated automatisms of the hy-
power losses cost. The investment presents a local minimum for brid loops enable to increase the SAIDI.
30 loops and then increases with the number of loops. If the
number of loops increases, then the total length of conductors C. Maximal DG Insertion Rate
required also increases, as does the investment. The disconti- Fig. 11 gives the maximum DG insertion rate obtained with
nuity of the investment comes from the discrete choice of the the Monte Carlo algorithm [17].
conductor whose cost is nonlinear. The actualized failure cost It varies from 38% to 67%. Other tests on real urban networks
increases with the number of loops because the rate of failure de- layout show that the current architecture cannot accept more
pends only on the length of conductor and not on the gauge. The than 25% of DG [32]. Building the secured feeder (c.f. Section
actualized power losses cost is not monotonous. This is also due II-A) on the same network enables hardly to reach 47%. This
to the discrete choice of conductors. A local minimum seems to curve is also nonmonotonous due to the same explanations as
be obtained for a number of loops between 30 and 100. for the costs and the reliability criteria.
B. Reliability Criteria D. Comparison Between the Hybrid Structure and the
Fig. 10 presents the evolution of the SAIDI and SAIFI when Secured Feeder
the number of loops increases. Both SAIDI and SAIFI have a Table I summarizes the results for the secured feeder and the
local minimum for 100 loops. Between 60 and 150 loops, the hybrid structure.
SAIDI is very good (lower than 3 min). The current SAIDI in The SAIDI and the EENS are much higher for the secured
France is in the mean of 15 min per year for urban feeder [15]. feeder than for the hybrid structure. The main reason is that the
ALVAREZ-HÉRAULT et al.: A NOVEL HYBRID NETWORK ARCHITECTURE TO INCREASE DG INSERTION IN ELECTRICAL DISTRIBUTION SYSTEMS 913

hybrid structure enables protecting each loop, and automatic Medium-term planning has to be studied to determine if the
switches increase the time of reenergizing consumers. Then current architectures are mutable towards the hybrid structure.
the reliability indexes are improved. Nevertheless, all the real The optimal investment planning could also be evaluated.
investment costs have not been taken into account. Protective
relays cost and the one related to the automation of switches ACKNOWLEDGMENT
have been approximated using the cost of a remotely controlled The authors would like to thank R. Jeannot (Schneider Elec-
switch. tric), A. Coiffier and F. Gorgette (ErDF), and G. Roupioz and S.
Concerning the different costs, the hybrid networks roughly Grenard (EDF R&D) for their very valuable advice and all the
reduce the global cost by reducing power losses costs and in- data provided during the study.
vestment costs. But as the real costs of protective relays have not
been taken into account, the global price is more or less equiv- REFERENCES
[1] IEEE Standard for Interconnecting Distributed Resources With Elec-
alent. tric Power Systems, IEEE Standards Coordinating Committee 21
To highlight this, if we take the example of the maximal (IEEE SCC21), Aug. 26, 2003.
reachable DG insertion rate for the secured feeder (47%), the [2] P. Abadie, Journal Officiel de la République française, arrêté du 23 avril
hybrid structure has a much better SAIDI of 4 min (against 12.8 2008 relatif aux prescriptions techniques de conception et de fonction-
nement pour le raccordement à un réseau public de distribution d’élec-
min for the secured feeder). Moreover, the global cost is about tricité en basse tension ou en moyenne tension d’une installation de
38 M euro (against 43 M euro for the secured feeder). Finally, if production d’énergie électrique, Apr. 25, 2008.
the DG scenario evolves, the DG insertion rate can be increased [3] N. Hadjsaid, J.-F. Canard, and F. Dumas, “Dispersed generation impact
on distribution networks,” IEEE Comput. Appl. Power, vol. 12, no. 2,
until 67% for the hybrid structure, whereas the secured feeder pp. 22–28, Apr. 1999.
is limited or would required expensive investment. [4] R. Caire, N. Retière, N. Hadjsaid, S. Martino, and C. Andrieu, “Per-
manent short circuit currents modification of low voltages distributed
generation on distribution network,” in Proc. IEEE 37th Int. Universi-
ties Power Engineering Conf. (UPEC 2002), Staffordshire, U.K., Sep.
9–11, 2002.
VI. CONCLUSION [5] S. Bacha, E. Lepelleter, J.-L. Schanen, R. Caire, D. Frey, and P. P.
Jeannin, “Short-circuit limitation thanks to a series connected VSC,”
The connection of distributed generation in the distribution in Proc. IEEE Applied Power Electronics Conf. Expo. APEC 2008,
network may have considerable impacts. They can be beneficial Austin, TX, Feb. 24–28, 2008.
[6] I. Etxeberria, U. Viscarret, M. Caballero, A. Rufer, and S. Bacha, “New
but also prejudicial (increasing of voltages, disturbance of pro- optimized PWM VSC control structures and strategies under unbal-
tection apparatus, etc.). Modern methods of planning could raise anced voltage transients,” IEEE Trans. Ind. Electron., vol. 54, no. 5,
their limits if the connection of DGs becomes massive. Then, it pp. 2902–2914, Oct. 2007.
[7] G. Celli, F. Pilo, and G. Pisano, “Meshed distribution networks to
is mandatory to include the DG parameters in long-term plan- increase the maximum allowable distributed generation capacity,” in
ning studies. Proc. CIGRE, 2005.
For this purpose, a novel architecture has been proposed as a [8] G. Celli, F. Pilo, G. Pisano, V. Allegranza, and R. Cicoria, “Distribu-
tion network interconnection for facilitating the diffusion of distributed
result of hybridization between the secured feeder and the loop. generation,” in Proc. 18th Int. Conf. Exhib. Electricity Distribution
The hybrid structure, robust to DGs interconnection, has main (CIRED 2005), Turin, Italy, Jun. 6–9, 2005.
feeders radially operated but enables a meshed operation of its [9] G. Celli, F. Pilo, G. Pisano, V. Allegranza, R. Cicoria, and A. Iaria,
“Meshed vs. radial MV distribution network in presence of large
loops. Autonomous breakers in each loop enable consumers to amount of DG,” in Proc. IEEE PES Power Systems Conf. Expo., 2004,
be rapidly reenergized in case of a fault. Oct. 10–13, 2004, vol. 2, pp. 709–714.
A heuristic algorithm to build the hybrid structure has been [10] S. Repo and H. Laaksonen, “A case study of voltage rise problem due
to a large amount of distributed generation on a weak distribution net-
proposed. The objective was to build equivalent PL product work,” in Proc. IEEE Bologna Power Tech Conf., Jun. 23–26, 2003,
areas minimizing the total length of conductors in order to guar- vol. 4.
anty the equality of customers with respect to the service con- [11] V. Calderaro, A. Piccolo, and P. Siano, “Maximizing DG penetration in
distribution networks by means of GA based reconfiguration,” in Proc.
tinuity. The DG parameter was insured by the loops of the hy- 2005 Int. Conf. Future Power Systems, Nov. 18, 2005.
brid structure. The results obtained on real data show that the [12] F.-A. Viawan, D. Karlsson, A. Sannino, and J. Daalde, “Protection
hybrid structure increases DGs connection and reliability. The scheme for meshed distribution systems with high penetration of dis-
tributed generation,” in Proc. PS’06 Power Systems Conf.: Advanced
presence of local minima gives the required number of loops re- Metering, Protection, Control, Communication, and Distributed Re-
garding the privileged index. sources, Clemson, SC, Mar. 14–17, 2006, pp. 99–104.
In this paper, the main purpose after increasing the DG inter- [13] S. Repo, “A method to increase the integration capacity of distributed
generation on weak distribution networks,” in Proc. 17th Int. Conf.
connection was to guaranty the equality of customers regarding Electricity Distribution, CIRED, May 2003, vol. 4.
the service continuity. Other parameters can also be taken into [14] N. Hadjsaid, M.-C. Alvarez-Hérault, R. Caire, B. Raison, J. Descloux,
account. If the planning operator prefers only minimizing the and W. Bienia, “Novel architectures and operation modes of distribu-
tion network to increase DG integration,” in Proc. IEEE PES General
total length of conductors, optimization methods exist to solve Meeting, Panel Session, Minneapolis, MN, Jul. 25–29, 2010.
this problem that turns out to be the Vehicle Routing Problem. [15] ERDF. [Online]. Available: http://www.erdfdistribution.fr.
Multiobjective functions could be used to consider several pa- [16] DGEMP—DIDEME, Coûts de Référence de la Production Électrique,
Dec. 2003.
rameters such as the risk of blindness or inopportune trips due to [17] B. Enacheanu, M.-C. Alvarez, B. Raison, R. Caire, W. Bienia, O. De-
the insertion of DGs in the distribution network. Furthermore, vaux, and N. HadjSaid, “Optimal meshed distribution network con-
some improvements can be added such as finding automatically figuration,” IREE (Int. Rev. Elect. Eng.), vol. 4, no. 5, pp. 957–966,
Sep./Oct. 2009.
the number of loops and feeder required for a DG insertion rate [18] P. Toth and D. Vigo, The Vehicle Routing Problem. Philadelphia, PA:
given target. SIAM, 2001, p. 363.
914 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 26, NO. 2, MAY 2011

[19] A. Schrijver, Combinatorial Optimization Polyhedra and Efficiency. Raphael Caire (M’04) received the DEA degree
New York: Springer, 2002. from INPGrenoble, Grenoble, France, in 2000.
[20] G. Clarke and J.-V. Wright, “Scheduling of vehicles from a central He was working in the power electronic field, in
depot to a number of delivery points,” Oper. Res., vol. 12, no. 4, pp. France at the Laboratoire de Génie Electrique de
568–581, Jul./Aug. 1964. Grenoble (LEG) and in the USA at the Center of
[21] N. Christofides, A. Mingozzi, and P. Toth, Combinatorial Optimiza- Power Electronic System (CPES). He is now an As-
tion. New York: Wiley, 1979. sociate Professor at Grenoble INP-ENSE3-G2Elab,
[22] B. E. Gillet and L. R. Miller, “A heuristic algorithm for the vehicle Saint Martin d’Hères, France. His research is
dispatch problem,” Oper. Res., vol. 22, no. 2, pp. 340–349, Mar./Apr. centered on the impacts, production control of
1974. dispersed generation on distribution system, and
[23] M. L. Fisher and R. Jaikumar, “A generalized assignment heuristic for critical infrastructures. He is with Grenoble Institute
the vehicle routing problem,” Revue Netw., Int. J., vol. 11, pp. 109–124, of Technology.
1981.
[24] J. E. Beasley, “Route first cluster-second methods for vehicle routing,”
J. Omega Ed. Elsevier, vol. 11, no. 4, pp. 403–408, Jan. 1983.
[25] R. F. Daganzo, C. F. Souleyrette, and R. Reginald, “Implementing ve- Bertrand Raison (M’03) was born in Béthune,
hicle routing models,” Transport. Res. Part B: Methodol., Elsevier, vol. France, in 1972. He received the M.S. and Ph.D.
24, no. 4, pp. 263–286, Aug. 1990. degrees in electrical engineering from the Grenoble
[26] G. Dueck and T. Scheurer, “Threshold accepting: A general purpose INP, Grenoble, France, in 1996 and 2000, respec-
optimization algorithm,” Revue J. Computat. Phys., Elsevier, vol. 90, tively.
no. 1, pp. 161–175, 1990. Since 2001, he has been with the Grenoble INP-
[27] J. A. G. Willard, Vehicle Routing Using r-Optimal Tabu Search. ENSE3-G2Elab as an Associate Professor. His gen-
London, U.K.: Manage. Sch., Imperial College, 1989. eral research interests are fault detection and localiza-
[28] J. H. Holland, Adaptation in Natural and Artificial Systems. Cam- tion in electrical systems. He is with Joseph Fourier
bridge, MA: MIT Press, 1975. University, Grenoble.
[29] H. Kawamura, M. Yamamoto, T. Mitamura, K. Suzuki, and A. Ohuchi,
“Cooperative search on pheromone communication for vehicle routing
problems,” IEEE Trans. Fundam., vol. E81-A, pp. 1089–1096, 1998.
[30] H. Ghaziri, “Solving Routing Problems by a Self-Organizing Map,” in
Artificial Neural Networks, T. Kohonen, K. Makisara, O. Simula, and Nouredine HadjSaid (SM’06) received Ph.D. and
J. Kangas, Eds. Amsterdam, The Netherlands: North-Holland, 1991. the “Habilitation à Diriger des Recherches” degrees
[31] M. Sakarovitch, Optimisation Combinatoire, Méthodes Mathéma- from the Grenoble Institute of Technology (Grenoble
tiques et Algorithmes, Programmation Discrète, Hermann, 1984. InP), Saint Martin d’Hères, France, in 1992 and 1998,
[32] M.-C. Alvarez-Hérault, R. Caire, B. Raison, N. HadjSaid, W. Bienia, respectively.
and J. Descloux, “Investigation of new distribution grid architecture for He served as an Assistant Professor at the Ecole
accommodating higher DG penetration rate,” in Proc. CIRED Work- Nationale Supérieure d’Ingénieurs Electriciens de
shop 2010, Lyon, France, Jun. 2010. Genoble and the Laboratoire d’Electrotechnique de
Grenoble from 1992 to 2000. He has been appointed
Marie-Cécile Alvarez-Hérault (S’10) was born in as a full Professor at the same institute since 2000.
Limoges, France, on April 5, 1983. She received the He is presently the General Director of a common
Engineering Diploma, the M.S. degree, and the Ph.D. research center between EDF, Schneider Electric, and INPGrenoble (IDEA) on
degree in electrical engineering from the Grenoble distributed generation and future electrical distribution systems.
INP, Grenoble, France, in 2007 and 2009.
Her research interest is new architectures and ex-
ploitation of distribution networks in the presence of
distributed generation. She is with Grenoble Institute Wojciech Bienia received the M.S. degree in
of Technology, Saint Martin d’Hères, France. mathematical applications from Silesian Polytechnic
School, Poland, in 1976 and the Ph.D. degree in
oriented matroids theory from University PARIS VI,
Paris, France, in 1985.
Since 1988, he has been an Associate Professor
Damien Picault was born in France in 1984. He re- at the Computer Science and Applied Mathematics
ceived the M.S. degree in electrical engineering from School of Grenoble (ENSIMAG)-Grenoble INP,
the Grenoble INP, Grenoble, France, in 2007 and is Saint Martin d’Hères, France, where he belongs
currently pursuing the Ph.D. degree in maximizing to the Sciences of Conception, Optimisation and
energy production of grid-connected photovoltaic Production (G-SCOP) group. His present research
systems. interests are operational research and combinatorial optimization. He is with
His main research interests are photovoltaic plant Grenoble Institute of Technology.
and power electronics modeling.