IEEE TRANSACTIONS ON SMART GRID, VOL. 8, NO.

3, MAY 2017 1211

A-Diakoptics for the Multicore Sequential-Time

Simulation of Microgrids Within Large
Distribution Systems
Davis Montenegro, Member, IEEE, Gustavo A. Ramos, Senior Member, IEEE, and Seddik Bacha, Member, IEEE

Abstract—This paper presents the A-Diakoptics methodology The methods for analyzing the Load flow in power sys-
(Diakoptics based on actors) for the high performance simulation tems were inspired by transmission systems (TS). These kind
of microgrids. Microgrids are clusters of loads and microsources of systems are meshed networks, assumed as balanced and
that operate within a single controllable system, which provides
electricity and heat to its local area even if they get separated have low X/R ratios; making them suitable for being analyzed
from the distribution system. Several commercial solutions are using methods based on Gauss-Seidel, Newton-Raphson and
addressing this topic using numerous methods; however, to keep its different versions [5], [6].
the computing time required to solve the system when the size On the other hand, the DS has special features that differ
of the system grows represents a major challenge. In this paper, from TS: These are radial or weakly meshed networks with
this challenge is addressed by separating the distribution grid
in independent subnetworks. Then, for the computational imple- non-transposed conducting lines, and the load is distributed
mentation, each subnetwork is assigned to an independent solver in an unbalanced way, requiring to consider all the phases to
using the actor model. As a result, the solution algorithm is perform the analysis.
composed of multiple parallel and concurrent solvers. These Additionally, the high penetration of non-conventional loads
exchange information using high-speed communication interfaces within the DS is a normal situation nowadays. Distributed
for reducing the computational burden without losing accuracy.
A-Diakoptics is an advanced algorithm that combines existing Energy Resources (DER) such as wind turbines (WT), pho-
methods for simulating large distribution systems; these meth- tovoltaic cell arrays (PV) and fuel cells change the con-
ods are upgraded to make them suitable for parallel processing ventional radial configuration of the grid [7]–[10]. These
and to cover the basic needs for the simulation of smart grid DERs grouped to create clusters of loads and microsources
applications. that operates within a single controllable system, which
Index Terms—Actor model, diakoptics, multicore process- provides electricity and heat to its local area even if
ing, parallel architectures, power system simulation, real time they get separated from the Distribution System in a
systems, reconfigurable architectures. microgrid [11].
As a result, with microgrids it is possible to enhance
I. I NTRODUCTION the local reliability, support local voltages, helps to miti-
HE SIMULATION of large distribution systems (DS) gate the effects of power quality disturbances, among other
T has become a major topic in the last decades [1], [2].
These simulations provide important information for sup-
benefits [12]. These features make the DS not adequate for
being analyzed by using the solving methods mentioned
porting management activities such as Advanced Distribution above [13].
Automation (ADA), which are coordinated from automated For analyzing the DS considering all the features mentioned
Energy Management Systems (EMS) [3], [4]. above, two reference frames have been used in the recent
decades: The phase frame and the sequence frame. The phase
Manuscript received March 6, 2015; revised March 20, 2015, August 9, frame refers to the techniques that consider all the calcula-
2015, and October 26, 2015; accepted December 7, 2015. Date of pub-
lication December 31, 2015; date of current version April 19, 2017. This tions based on using the 3 phase a-b-c [14], [15]. Instead,
work was supported in part by the Departamento Administrativo de Ciencia y the sequence frame separates the DS in three phasor systems,
Tecnología de Colombia COLCIENCIAS under Convocatoria 528-2011, and which are solved separately and then are superposed to find
in part by the Universidad Santo Tomás Colombia. Paper no. TSG-00261-
2015. the total solution [16].
D. Montenegro is with the Electrical and Electronics Engineering However, modern computing architectures are multicore,
Department, Universidad de los Andes, Bogotá 111711, Colombia, and also which allows to implement several independent programs
with the G2ELab-INP, Université Joseph Fourier, Grenoble 38031, France
(e-mail: [email protected]). working in parallel to solve a common problem [17]. These
G. A. Ramos is with the Electrical and Electronics Engineering Department, technological advances represent an opportunity to improve
School of Engineering, Universidad de los Andes, Bogotá 111711, the existing solvers, which are normally implemented using
Colombia (e-mail: [email protected]).
S. Bacha is with the Electrical Engineering Department, G2ELab- sequential programming, by using parallel programming on
INP, Université Joseph Fourier, Grenoble 38031, France (e-mail: a single computer.
[email protected]). It has been proven that by distributing tasks within a set
Color versions of one or more of the figures in this paper are available
online at http://ieeexplore.ieee.org. of parallel solvers helps to analyze the power flow prob-
Digital Object Identifier 10.1109/TSG.2015.2507980 lem in low computing times [18]. Techniques such as the
1949-3053 c 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
1212 IEEE TRANSACTIONS ON SMART GRID, VOL. 8, NO. 3, MAY 2017

Bordered Block Diagonal Matrix (BBDM) have been used for system load flow becomes determined as follows:
this purpose; nevertheless, they have been conceived based on
Newton-Raphson solvers and focused in TS within a computer ⎡ ⎤ ⎡ ⎤−1 ⎡ ⎤
V1 Y11 0 ··· 0 I1
network. ⎢ V2 ⎥ ⎢ 0 Y22 ··· 0 ⎥ ⎢ I2 ⎥
⎢ ⎥ ⎢ ⎥ ⎢ ⎥
Parallel programming within a single PC can lead ⎢ .. ⎥ = ⎢ .. .. .. .. ⎥ ⎢ .. ⎥ (2)
to improve the computing performance of fast simu- ⎣ . ⎦ ⎣ . . . . ⎦ ⎣ . ⎦
lators and Real-Time-Hardware In the Loop simulators Vn 0 0 · · · Ynn In
(RT-HIL) [19], [20], which are commonly used as the base
for EMS and for testing new power technologies in realistic The ways the parts of (2) interact are defined by the different
situations [21]. approaches within this frame. These approaches can handle
In this paper, this challenge is addressed by separating the DG and diverse network topologies in balanced or unbalanced
distribution grid in independent sub networks. Then, for the conditions.
computational implementation each sub network is assigned The forward and backward sweep algorithm [24], [25]
to an independent solver using the actor model. counts with three approaches: The current summation
As a result, the solution algorithm is composed by multi- approach [26], the power summation approach [27] and the
ple parallel and concurrent solvers that exchange information admittance summation approach [28]. These approaches are
using high speed communication interfaces, thus reducing the inspired in Kirchhoff’s Current Law (KCL) and Kirchhoff’s
computational burden without losing accuracy. This is possi- Voltage Law (KVL); the aim is to determinate the grid voltage
ble by using the existing multicore computing architectures, by injecting currents and compensate them until reaching con-
which allows the processing of parallel and concurrent tasks vergence. These have also demonstrated to be suitable for DG
independently. specially the power summation approach; where the genera-
This method is an advanced algorithm that combines exist- tors are replaced by power injection units and the convergence
ing methods for simulating large DS; these methods are is evaluated in terms of power [29]. However, they represent
upgraded to make them suitable for parallel processing, and a considerable computational burden when the topology of the
to cover the basic needs for the simulation of smart grid grid is not strictly radial.
applications. On the other hand, the compensation algorithm, which is
This method is implemented in a customized version of based in forward and backward sweep algorithm, proposes
EPRI’s OpenDSS [22] complemented with advanced features. to find compensation currents to complement the injection
This new software is called DSSim-XX and counts with current vector [30]. With this compensated vector the com-
PC [23] and Real-Time (RT) versions. This simulator was pensation algorithm looks for opening loops for guaranteeing
developed using NI LabVIEW. the radial topology of the system to solve [31]. This algorithm
This article is divided in four sections: 1) A brief review can handle DG and weakly meshed networks.
about the solution methods used in the last years for DS and Other algorithms like Implicit ZBUS Gauss [14], Modified
the RT-HIL approach. 2) The A-Diakoptics method. 3) Results Newton [32] and Direct methods such as Bus Injection to
when applying A-Diakoptics to medium-scale and large-scale Branch Current/Voltage (BIBC/BICV) [33], are focused in the
realistic DS and finally 4) Conclusions. factorization of the Y matrix for obtaining a block triangular
form (BTF). This factorization is the base for solving faster the
power system and reach convergence with less iterations.
II. T HE S OLUTION M ETHODS FOR L OAD The convergence criterion can be based in the voltages of
F LOW A NALYSIS IN DS the system, also the currents and power can be considered.
As mentioned in the previous section, there are currently These algorithms can handle DG, weakly meshed and meshed
two frames for load flow analysis in DS. These are com- networks.
posed by several iterative methods looking for covering all
the features and needs for smart grid studies. Both frames are
described as follows: B. Power Flow Analysis Based on Sequence Frame
Instead of the phase frame, the sequence frame decomposes
the unbalanced DS into its positive, negative and zero sequence
A. Power Flow Analysis Based on Phase Frame networks for being solved in a decoupled form. The advantage
The phase frame refers to the algorithms that consider three of this decomposition is that the positive sequence network is
phases power flow analysis for DS. In this frame the size of balanced, which allow using the single phase Newton-Raphson
the Y matrix that describes the power system is defined by the method for solving the network [34], [35].
following expression: Finally, the decoupled solutions are coupled again and this
m process keeps until reaching convergence, which is performed
#Rows Y = Nn (1) by evaluating the voltage instead of power.
n=1
The advantage about this method is that it reduces the
In (1) m corresponds to the number of buses of the system computational burden by simplifying the problem. The decou-
and Nn the number of nodes of the bus. The nodes refer to the pled solution allows also the distributed analysis of the power
number of phases connected to the bus m. Then, the power system. This method can handle DG and several topologies.
MONTENEGRO et al.: A-DIAKOPTICS FOR THE MULTICORE SEQUENTIAL-TIME SIMULATION OF MICROGRIDS 1213

The frames presented above consider the system as a single

block; this situation represents several disadvantages related
with the performance of the simulator:
1) When the number of nodes of the DS increases also
the computing time; this makes difficult to guarantee
deterministic computing times for RT-HIL simulation.
2) The simulation of sophisticated control devices interact-
ing within the DS can add an important computational
burden; affecting the performance of the simulation.
3) Topological changes represent important computation
time for being considered within the simulation. Fig. 1. Decomposition of the interconnected network (Cn=Contours).
Depending on the number of phases of a switch/line,
the number of cells to update when they trip can be methods [38]. These were designed for TS and their appli-
significant. cations include economic dispatch, power flow analysis and
4) These methods are not suitable for multicore/parallel transient stability analysis. Diakoptics was formally proposed
processing using nowadays computing architectures. by Kron in the 55’s [36] and since then several authors have
proposed different interpretations for the algorithm.
One of these interpretations is Z-Diakoptics, which was
III. T HE A-D IAKOPTICS A LGORITHM
used for performing power flow analysis using a distributed
The algorithm Diakoptics based on Actors (A-Diakoptics) network of computers. This was proposed by Happ in the
is an algorithm that combines Diakoptics [36] and the actor 70’s [39] and proposes the decomposition of the system using
model [37]. The aim of this algorithm is to make DS con- equivalent impedance matrixes. However, when the computing
ception suitable for multicore processing and improve the architectures became more powerful the piecewise methods
simulation performance. This improvement covers standard became irrelevant; this because one computer was enough for
and Real-Time computing architectures. For evaluating the performing power flow analysis on TS.
performance of this algorithm, EPRI’s OpenDSS is used. The interesting about this method is that it allows con-
sidering primitive sub-network as independent systems. This
A. Why EPRI’s OpenDSS? is possible by decomposing the interconnected network into
primitive and orthogonal networks.
OpenDSS is an open source simulator that includes sev-
Primitive networks are the sub-networks formed when sep-
eral features for supporting smart grid studies. This simulator
arating the interconnected system; orthogonal networks are
counts with a COM interface (OpenDSSEngine) for exchang-
those that interconnect the primitive networks. The primitive
ing data with external software [9].
networks do not consider the frontier effects produced by the
OpenDSS was selected from other available open source
interconnection links, they are declared like if the other net-
platforms because of the following reasons:
works did not exist. On the contrary, the orthogonal networks
1) The code is well coded. The code was conceived in term
include the information about the network view from the link
of classes and objects, making easier the code reading
branch.
and modification.
As result the interconnected is divided in 4 subsets as
2) The COM interface is a good alternative for interfac-
shown in Fig. 1. Matrix ZTT refers to the trees matrix; this
ing external software; however, to not to compromise
contains the inverted Y matrixes of the isolated subsystems
the simulator performance it must be used using early
when tearing the interconnected system [40]. Matrix ZCC
binding connection methods.
refers to the connections matrix; this is built by combining
3) Includes a good number of models and control devices
the contours matrix (also called tensors) with ZTT and the
and there are several study cases documented.
links between the subsystems [41]. ZCT and ZTC are comple-
4) Supports all steady-state analysis commonly performed
mentary matrixes obtained with partial components of ZCC .
for utility DS planning. This software is the evolution
As result, the expression presented in (2) is transformed as
of harmonics solvers rather than of power flow solvers,
follows [42], [43]:
which simplifies to solve the power flow problem and
⎧⎡ ⎤ ⎡ ⎤⎫
⎨ EA + eA
⎪
adds new types of analysis [22]. IA ⎪
⎢ ⎥    ⎢ IB ⎥⎬
5) The solver of OpenDSS uses Modified Newton tech- ET ⎢ EB + eB ⎥  ⎢ ⎥ I0
niques for performing power flow analysis. This makes ⎩⎢
⎪ .. ⎥ = ZTT  ZTC ⎢ . ⎥⎪
⎢ .. ⎥⎭
⎢ ⎥ (3)
of OpenDSS a good candidate for applying A-Diakoptics .  
   ZCT ZCC
and to evaluate how its performance gets improved. e c e +e c cc I } i c c

The terms for obtaining ZCC , ZTC and ZCT are:

B. Diakoptics and the Piecewise Methods
ZCC = CT ZTT C + ZLLD (4)
Diakoptics is a method used for the decomposition of large-
scale networks in a set of independent sub-networks. This ZCT = CT ZTT (5)
method and its variations are well known as the piecewise ZTC = ZTT C (6)
1214 IEEE TRANSACTIONS ON SMART GRID, VOL. 8, NO. 3, MAY 2017

TABLE I
G ENERALIZED D IAKOPTICS A LGORITHM FOR
S OLVING A P IECEWISE P OWER S YSTEM

Fig. 2. Interactions between orthogonal and primitive networks to find the

total solution.

Fig. 3. Evolution of vector I0 when the simulation is performed.

Fig. 4. Actor framework for the A-Diakoptics method.
ZLLD is the matrix describing the link branch as a set of
diagonal impedances; the size if this matrix is equal to the
As shown in Fig. 3, the components of I0 change in time as
number of phases of the link branch. This approach assumes
product of the compensation currents found when solving the
that the subsystems are radially attached, which led to consider
independent subsystems. As consequence, Equation (7) must
the absence of voltage sources in the link branches making
be re-evaluated in time as follows:
ec equal to 0. Considering this the solution of the power
−1
system is: ET = ZTT I0(t1) − ZTC ZCC ZCT I0(t3) (8)
−1
ET = ZTT I0 − ZTC ZCC ZCT I0 = E0 + E1 (7) In Fig. 3, the Childs are the programs that solve each sub-
The left part of (7) refers to the solution of the independent system independently. The parent refers to the program that
subsystems and the right part, the complementary solution. sends data to the Children and manages the obtained results.
The vector I0 is the currents injection vector; this includes the The algorithm for implementing Equation (8) is presented in
currents injected by the Thévenin equivalents of slack buses TABLE I.
and voltage sources [44]. This vector also includes the currents As can be seen in this algorithm, sequential programming
injected by other current sources to the system. is not enough for its implementation. Several parts of the
The solution of each subsystem can be found by using any algorithm must be implemented in parallel and executed con-
method of those presented in Section II, which allows to make currently to grant an improved performance; otherwise, the
them compatible with parallel processing. In this work, the application this could be not as good as expected. This need
KLU Solver and the iterative solution proposed by OpenDSS leads to consider sophisticated programing frameworks for
is implemented on the child actor [45]; as a result, each child using the existing multicore computing architectures, hence
actor behaves as an independent OpenDSS solver. Equation (7) the actor model is proposed for covering this need.
can be interpreted as follows: First, each subsystem must be
solved independently for finding E0 and complement the cur- C. The Actor Model as Framework
rents vector I0 . Then, with this updated vector the terminals of The actor model was proposed by Hewitt in the 70s [46].
the link branches can be excited for obtain the complementary Later this model was used for RT computation and for
current ic . integrate heterogeneous computing systems [47].
With this current the complementary voltage E1 can be An actor is a Queue-Driven State Machine (QDSM) that
found and added to E0, which was calculated previously. These exchanges data with other actors by using messages. The mes-
interactions in time are illustrated in Fig. 2. sages are sent by using dedicated queues, which are addressed
According to the interpretation described above I0 is not to guarantee the high speed communication between these.
the same in time for finding the total solution of the system. Actors can be defined as objects within a virtual environment.
MONTENEGRO et al.: A-DIAKOPTICS FOR THE MULTICORE SEQUENTIAL-TIME SIMULATION OF MICROGRIDS 1215

Fig. 5. RT-HIL version of the simulator DSSim-RT.

As objects, they can be cloned automatically and executed for building ZTT . These requirements could be the maximum
independently in parallel concurrently [48]. size of the primitive networks, the existence of switches or
For creating actors these must consider several parts; these a weighted combination of the above. These requirements
include the number of states that define the behavior of define the number of child actors.
the actor in time. The formal definition of an actor is as Then the parent calculates the orthogonal matrix ZCC and
follows: the matrixes ZCT and ZTC . After these operations the parent
actor launches the child actors for solving the subsystems.
A = (S, T, I, Q) (9)
The child actors solve each subsystem independently and
Where S is the set of the actor states, T is the set of returns to the parent the currents and voltages calculated. Then
actions of the actor, I is the set of initial/active states and the parent calculates the ic and E1 using the algorithm in
Q is the set of quiescent states (normally I is the complement TABLE I. The functional and hierarchical schemas of the actor
of Q). For this implementation, it is used the actor framework system are shown in Fig. 4.
available on NI LabVIEW [49], which is the programming It is because of the multicore processors that it is possible
language used for developing DSSim-XX [23]. This soft- to execute parallel task within a single computer. Each actor
ware uses the structure of the solver provided by EPRI’s in is handled by the local operating system (OS) for being exe-
OpenDSS. cuted in a specific processor or thread. However, for RT-HIL
In this framework the messages between actors can be high purposes the assignation of hardware resources is controlled
coupled, low coupled or zero coupling. For A-Diakoptics the by DSSim-XX. In this case the assignation of resources is
zero coupled model was chosen. With zero coupling messag- made within a homogeneous computing environment; due to
ing the messages can be used for sending different types of the multiple cores are part of a single processor.
messages; no matter if the states of the destination actor have But this method can be also applied to heterogeneous com-
changed. puting environments, turning the multiple computing devices
This structure is composed of two elements: The first is into actors. That is the case of the RT-HIL version of the sim-
a numeric index that describes the type of message; the second ulator; in this version the devices for acquiring and generating
is a variant element for sending any type of data. This way, power signals are external to the simulation unit.
depending on the type of message the remote actor transforms As an example of this concept, the operating principle of
the variant in the type of data for being used locally. the DSSim-RT version is shown in Fig. 5, which uses standard
This implementation has three hierarchical levels: The PCs and NI cRIO hardware.
launcher, the parent actor and the child actors. The launcher Each external actor communicates with the others by using
starts the parent actor by sending the system’s Y matrix, the communication networks like TCP/Ethernet; this way, the
initial excitation vector (currents), the location of loads and tasks are distributed for reaching deterministic computing
the type (generator, load to ground, etc.). These values are times. The external actors can be configured like signal gen-
requested to the OpenDSSEngine. erators, signal acquisition devices, communication handlers,
The parent actor separates the Y Matrix and creates the among other functionalities to interact with the simulated
primitive networks depending on the launcher requirements system.
1216 IEEE TRANSACTIONS ON SMART GRID, VOL. 8, NO. 3, MAY 2017

Fig. 7. IEEE 8500 nodes system (8561 nodes) modeled on DSSim-PC/RT.

Fig. 6. EPRI’s circuit 5 (3437 nodes) modeled on DSSim-PC/RT. TABLE II

C HARACTERISTICS OF THE S IMULATED S YSTEMS

IV. R ESULTS
To evaluate the proposed method, two power systems are
implemented in the DSSim-PC and DSSim-RT. These simu-
lations are compared with the standard version of OpenDSS TABLE III
to verify the performance in all the applications. R ESULTS ACHIEVED
The system 1 is a medium-scale power system. This power
system is the EPRI’s circuit 5 and is shown in Fig. 6. The
system 2 is a large-scale power system. This system is the
IEEE 8500 node system and is shown in Fig. 7.
The features of these systems are shown in TABLE II. For
the separation process the algorithm is weighted 70% switches
and 30% the size of each area. The maximum size of each
area is configured to be the 15% of the system’s Y Matrix.
This way, the separation process will group nodes into areas
where the maximum number of nodes is limited (1000 nodes The second architecture is an RT architecture for testing
for EPRI’s circuit 5 and 1650 nodes for IEEE 8500); however, the RT performance. This architecture has an Intel CoreTM
this distribution can consider more separations to include all i7-4810MQ Processor 3.4 GHz (4 cores), 16 GB RAM and
the possible topology changes (100% switches). OS Windows 7 embedded. The OS is dedicated to DSSim-XX
This is an important consideration for the simulation pro- and only the strictly necessary applications were running at
cess, if the separation process considers 100% switches, means the moment of the test. The results of each test are shown in
that the system will separate the network in N+1 primitive TABLE III.
networks, being N the number of switches. As a result, when The simulation time is measured without considering the
modifying the state of these link branches within ZCC the new time required for compiling the system. These results reveal
topology can be solved without needing to update the orthog- a significant reduction on the total processing time and mem-
onal matrixes. Nevertheless, the solution time should not be ory used. In both cases (PC and RT) the separation of
affected, but the amount of memory used will increase (more a big matrix into small independent subsystems represents
child actors). a reduction of the memory needed for describing the network.
Two hardware architectures were used for the testing stage. However, in the case of the PC the results can vary depending
The first is a conventional architecture for testing the PC on the number of external applications running.
version of DSSim. This architecture has an Intel CoreTM The simulation tests were complemented by adding DERs,
i5 processor 2.4 GHz (2 cores), 8 GB RAM and OS Windows the aim is to evaluate the behavior of the microgrid in time
7 professional. Software such as antivirus, firewall, among when this is configured in islanding mode. In this case photo-
others, was running at the moment of the test. voltaic systems (PVSystem) are added to the IEEE 8500 node
MONTENEGRO et al.: A-DIAKOPTICS FOR THE MULTICORE SEQUENTIAL-TIME SIMULATION OF MICROGRIDS 1217

Fig. 9. Representation of link branches (switches) using the matrix ZCC

(Cn = Contours).

Since then, the microgrid supplies only the local load in

the time period when the PVSystem variables allow it. This
is illustrated on Fig. 8.
Fig. 8. DER simulation in DSSim-XX. Graph generated with DSSim-XX. In Fig. 8 two buses are shown: The bus SX3727712A and
the bus m1026810; both buses are three phase and are in differ-
TABLE IV ent locations in the system IEEE 8500. The bus m1026810 is
VOLTAGES C ALCULATED BY E ACH P LATFORM AND E RROR
out of the microgrid and near to the substation.
The accuracy of the solution is not affected as shown in
TABLE IV; in this table, the values calculated for differ-
ent buses in different locations within the circuit are shown.
These locations were selected randomly and the highest error
achieved is 1E-5, which demonstrates that the accuracy of
OpenDSS is not significantly affected.
Another result is the reduction in the number of cells
required for representing link branches. As shown in Fig. 9,
a simple system with two islands is separated by using a three
phase switch. The colored cells represent the places different
from 0 and the cells marked with “X” represent the places
required for representing the link branch (switch).
As can be seen in Fig. 9 when using traditional methods
(left side) 12 cells are required for representing the switch; on
the other hand, with A-Diakoptics (right side) only three cells
are required for representing the state of the switch, thus reduc-
ing the computational time required for updating the topology
of the system.

system and the simulation is performed in sequential-time. At V. C ONCLUSION

the sequential time mode load profiles are linked to the loads A-Diakoptics is an advanced algorithm that combines exist-
for changing its power consumption in time. On the other ing methods for simulating large DS; these methods are
hand, PVSystems deliver the power according to the irradi- upgraded to make them suitable for parallel processing, and
ance profile, the temperature profile and the fixed efficiency to cover the basic needs for the simulation of smart grid
in OpenDSS [50]. This way, the simulation is time-based for applications.
performing control actions and to evaluate the system behavior. With this algorithm the four challenges proposed in
The behavior in time of each PVSystem and the loads is Section II have been addressed as follows:
recalculated each iteration by the local child actor. The aim is 1) By distributing tasks using multicore architectures the
to reduce the computational burden of the solution algorithm simulation time can be preserved in function of the
by distributing tasks. hardware architecture and the scale of the DS. With
To handle the behavior of the PVSystem do not represent an this algorithm the traditional power flow analysis can
important computation time with this methodology; the total be taken to a multicore computing architecture. This
solving time is not affected by this process. will improve the simulation performance by using the
A photovoltaic plant of 750 kW is added to the node existing computing power.
SX3727712A of the system IEEE 8500 Nodes [51]. The load 2) Because the simulation is distributed within homo-
in the island consumes only 650 kW and the rest of the power geneous computing environments, complex tasks can
is delivered to the DS. For the first 24 hours the microgrid at be integrated into it. So, sophisticated controllers can
the node SX3727712A is connected to the system; then, the be integrated into the simulation without affecting the
microgrid is separated when the link switch trips at 1:00 AM. performance of it.
1218 IEEE TRANSACTIONS ON SMART GRID, VOL. 8, NO. 3, MAY 2017

3) Due to the distributed representation of the DS, the topo- [12] J. Eto et al., “Overview of the CERTS microgrid laboratory test bed,” in
logical changes can be considered without affecting the Proc. CIGRE/IEEE PES Joint Symp. Integr. Wide-Scale Renew. Resour.
Power Del. Syst., Calgary, AB, Canada, 2009, p. 1.
simulation computing time. The number of cells to be [13] K. Balamurugan and D. Srinivasan, “Review of power flow studies
updated is drastically reduced. on distribution network with distributed generation,” in Proc. IEEE
4) With A-Diakoptics the traditional power flow meth- 9th Int. Conf. Power Electron. Drive Syst. (PEDS), Singapore, 2011,
pp. 411–417.
ods can handle multicore computing architectures; an [14] T.-H. Chen, M.-S. Chen, K.-J. Hwang, P. Kotas, and E. A. Chebli,
advance that can be used for incorporating advanced “Distribution system power flow analysis—A rigid approach,” IEEE
features into existing simulation platforms such as Trans. Power Del., vol. 6, no. 3, pp. 1146–1152, Jul. 1991.
[15] M.-S. Chen and T.-H. Chen, “Application of three-phase load flow to
OpenDSS. power system distribution automation,” in Proc. Int. Conf. Adv. Power
The step times reported in this paper are close to the times Syst. Control Oper. Manage. (APSCOM), vol. 2. Hong Kong, 1991,
reported by existing simulation tools for Real-Time, which pp. 472–478.
[16] M. Z. Kamh and R. Iravani, “Unbalanced model and power-flow analysis
makes this development competitive. of microgrids and active distribution systems,” IEEE Trans. Power Del.,
Diakoptics is itself a solution method; however, it can vol. 25, no. 4, pp. 2851–2858, Oct. 2010.
be also used to upgrade existing solution methods into par- [17] J. Diaz, C. Munoz-Caro, and A. Nino, “A survey of parallel program-
ming models and tools in the multi and many-core era,” IEEE Trans.
allel processing. Additionally, the logic proposed using the Parallel Distrib. Syst., vol. 23, no. 8, pp. 1369–1386, Aug. 2012.
actor model allows to handle each primitive network with an [18] M. Shahidehpour and Y. Wang, Communication and Control in Electric
independent solver within the same computer. Power Systems: Applications of Parallel and Distributed Processing.
Hoboken, NJ, USA: Wiley, 2004.
When mentioning OpenDSS as the improved base sys- [19] D. Z. Fang and Y. Xiaodong, “A new method for fast dynamic simulation
tem, the potential users of the presented development are of power systems,” IEEE Trans. Power Syst., vol. 21, no. 2, pp. 619–628,
also mentioned, these include utilities (R&D), which can use May 2006.
[20] Y. Deng, S. Foo, and H. Li, “Real time simulation of power flow control
this platform for planning, testing, and training, among other strategies for fuel cell vehicle with energy storage by using real time
uses. In fact, recently ANEEL (National Agency of Electric digital simulator (RTDS),” in IPEMC IEEE 6th Int. Power Electron.
Energy-Brazil) has proposed OpenDSS as the default platform Motion Control Conf., Wuhan, China, 2009, pp. 2323–2327.
[21] V. Jalili-Marandi, F. J. Ayres, E. Ghahremani, J. Belanger, and
to calculate losses in Brazil, which shows the potential uses V. Lapointe, “A real-time dynamic simulation tool for transmission and
of the improvement that we are proposing. distribution power systems,” in Proc. IEEE Power Energy Soc. Gen.
The A-Diakoptics algorithm is not yet implemented in the Meeting (PES), Vancouver, BC, Canada, 2013, pp. 1–5.
[22] R. C. Dugan and T. E. McDermott, “An open source platform for col-
DSSim-PC version available on the internet. This will be laborating on smart grid research,” in Proc. IEEE Power Energy Soc.
implemented in the source code of OpenDSS for making it Gen. Meeting, San Diego, CA, USA, 2011, pp. 1–7.
available for the users of OpenDSS and DSSim-PC. [23] D. Montenegro. (Jul. 1, 2013). DSSim-PC, Electrical
Distribution Network Simulator for PC. [Online]. Available:
http://sourceforge.net/projects/dssimpc/?source=navbar
[24] W. H. Kersting and R. C. Dugan, “Recommended practices for dis-
R EFERENCES tribution system analysis,” in Proc. IEEE PES Power Syst. Conf.
Expo. (PSCE), Atlanta, GA, USA, 2006, pp. 499–504.
[1] C. Dufour, H. Le-Huy, J. C. Soumagne, and A. El Hakimi, “Real-time
[25] W. H. Kersting, Distribution System Modeling and Analysis. Boca Raton,
simulation of power transmission lines using Marti model with opti-
FL, USA: CRC Press, 2001.
mal fitting on dual-DSP card,” IEEE Trans. Power Del., vol. 11, no. 1,
[26] G. W. Chang, S. Y. Chu, and H. L. Wang, “An improved back-
pp. 412–419, Jan. 1996.
ward/forward sweep load flow algorithm for radial distribution systems,”
[2] W. Ren et al., “Interfacing issues in real-time digital simulators,” IEEE IEEE Trans. Power Syst. vol. 22, no. 2, pp. 882–884, May 2007.
Trans. Power Del., vol. 26, no. 2, pp. 1221–1230, Apr. 2011. [27] S. Ghosh and D. Das, “Method for load-flow solution of radial distri-
[3] P. Aristidou, D. Fabozzi, and T. Van Cutsem, “Dynamic simulation bution networks,” IEE Proc. Gener. Transm. Distrib., vol. 146, no. 6,
of large-scale power systems using a parallel schur-complement-based pp. 641–648, Nov. 1999.
decomposition method,” IEEE Trans. Parallel Distrib. Syst., vol. 25, [28] D. Rajičić and R. Taleski, “Two novel methods for radial and weakly
no. 10, pp. 2561–2570, Oct. 2014. meshed network analysis,” Elect. Power Syst. Res., vol. 48, no. 2,
[4] F. Zavoda, “Advanced distribution automation (ADA) applications and pp. 79–87, 1998.
power quality in smart grids,” in Proc. China Int. Conf. Elect. Distrib., [29] S. M. Moghaddas-Tafreshi and E. Mashhour, “Distributed generation
Nanjing, China, 2010, pp. 1–7. modeling for power flow studies and a three-phase unbalanced power
[5] B. Stott, “Review of load-flow calculation methods,” Proc. IEEE, vol. 62, flow solution for radial distribution systems considering distributed
no. 7, pp. 916–929, Jul. 1974. generation,” Elect. Power Syst. Res., vol. 79, no. 4, pp. 680–686, 2009.
[6] N. H. Dandachi and B. J. Cory, “Network flow methods applied to [30] Y. Zhu and K. Tomsovic, “Adaptive power flow method for distribution
power system problems—A survey paper,” in Proc. Int. Conf. Adv. systems with dispersed generation,” IEEE Trans. Power Del., vol. 17,
Power Syst. Control Oper. Manage. (APSCOM), vol. 2. Hong Kong, no. 3, pp. 822–827, Jul. 2002.
1991, pp. 621–626. [31] C. S. Cheng and D. Shirmohammadi, “A three-phase power flow method
[7] T. S. Ustun, C. Ozansoy, and A. Zayegh, “Distributed energy for real-time distribution system analysis,” IEEE Trans. Power Syst.,
resources (DER) object modeling with IEC 61850–7–420,” in Proc. 21st vol. 10, no. 2, pp. 671–679, May 1995.
Austral. Univ. Power Eng. Conf. (AUPEC), 2011, pp. 1–6. [32] P. A. N. Garcia, J. L. R. Pereira, S. Carneiro, Jr., V. M. da Costa,
[8] R. F. Arritt and R. C. Dugan, “Distribution system analysis and the and N. Martins, “Three-phase power flow calculations using the current
future smart grid,” IEEE Trans. Ind. Appl., vol. 47, no. 6, pp. 2343–2350, injection method,” IEEE Trans. Power Syst., vol. 15, no. 2, pp. 508–514,
Nov./Dec. 2011. May 2000.
[9] R. C. Dugan, R. F. Arritt, T. E. McDermott, S. M. Brahma, and [33] J. H. Teng and C.-Y. Chang, “A novel and fast three-phase load flow
K. Schneider, “Distribution system analysis to support the smart grid,” in for unbalanced radial distribution systems,” IEEE Trans. Power Syst.,
Proc. IEEE Power Energy Soc. Gen. Meeting, Minneapolis, MN, USA, vol. 17, no. 4, pp. 1238–1244, Nov. 2002.
2010, pp. 1–8. [34] K. L. Lo and C. Zhang, “Decomposed three-phase power flow solu-
[10] Z. Ning et al., “Planning pumped storage capacity for wind power tion using the sequence component frame,” IEE Proc. C Gener. Transm.
integration,” IEEE Trans. Sustain. Energy, vol. 4, no. 2, pp. 393–401, Distrib., vol. 140, no. 3, pp. 181–188, May 1993.
Apr. 2013. [35] M. Abdel-Akher, K. M. Nor, and A. H. A. Rashid, “Improved three-
[11] R. H. Lasseter, “Microgrids,” in Proc. IEEE Power Eng. Soc. Winter phase power-flow methods using sequence components,” IEEE Trans.
Meeting, vol. 1. New York, NY, USA, 2002, pp. 305–308. Power Syst., vol. 20, no. 3, pp. 1389–1397, Aug. 2005.
MONTENEGRO et al.: A-DIAKOPTICS FOR THE MULTICORE SEQUENTIAL-TIME SIMULATION OF MICROGRIDS 1219

[36] G. Kron, “Detailed example of interconnecting piece-wise solutions,” Davis Montenegro (M’11) received the degree from
J. Franklin Inst., vol. 259, no. 4, pp. 307–333, 1955. the Universidad Santo Tomás, Bogotá, Colombia, in
[37] C. Hewitt and H. Barker, “Actors and continuous functionals,” Lab. 2004, and the M.Sc. degree from the Universidad
Comput. Sci., Mass. Inst. Technol., Cambridge, MA, USA, Tech. Rep. de los Andes, Bogotá, in 2012, both in electronic
MIT/LCS/TR-194, 1977. engineering. He is currently pursuing the Ph.D.
[38] G. Kron, Diakoptics: The Piecewise Solution of Large-Scale Systems. degree with the Universidad de los Andes and the
London, U.K.: Macdonald, 1963. Institut National Polytechnique de Grenoble, France.
[39] H. H. Happ, Diakoptics and Networks, vol. 1, 1st ed. New York, NY, For this period, he has a work license from the
USA: Elsevier, 1971. Universidad Santo Tomás, where he is a Professor
[40] H. H. Happ, Piecewise Methods and Applications to Power Systems. with the Department of Electronic Engineering. He
Hoboken, NJ, USA: Wiley, 1980. is a Specialist in automation of industrial processes
[41] H. H. Happ, “Orthogonal networks,” IEEE Trans. Power App. Syst., in 2006.
vol. PAS-85, no. 3, pp. 281–294, 1966.
[42] H. H. Happ, “Z diakoptics—Torn subdivisions radially attached,” IEEE
Trans. Power App. Syst., vol. PAS-86, no. 6, pp. 751–769, Jun. 1967.
[43] H. H. Happ, “Special cases of orthogonal networks—Mesh and nodal
networks,” IEEE Trans. Power App. Syst., vol. PAS-87, no. 1, pp. 53–66,
Jan. 1968. Gustavo A. Ramos (M’04–SM’14) received the
[44] H. E. Brown, G. K. Carter, H. H. Happ, and C. E. Person, “Power flow degree from the Universidad Nacional, Manizales,
solution by impedance matrix iterative method,” IEEE Trans. Power App. Colombia, in 1997, and the M.Sc. and Ph.D.
Syst., vol. 82, no. 65, pp. 1–10, Apr. 1963. degrees from the Universidad de los Andes,
[45] T. A. Davis and E. P. Natarajan, “Algorithm 907: KLU, a direct Bogotá, Colombia, in 1999 and 2008, respec-
sparse solver for circuit simulation problems,” ACM Trans. Math. Softw., tively, all in electrical engineering. He is cur-
vol. 37, no. 3, Sep. 2010, Art. ID 36. rently an Associate Professor with the Department
[46] C. Hewitt, “Viewing control structures as patterns of passing of Electrical Engineering, School of Engineering,
messages,” Artif. Intell. Lab., Mass. Inst. Techonol., Cambridge, Universidad de los Andes, Bogotá.
MA, USA, Tech. Rep. AIM-410, Dec. 1976. [Online]. Available:
http://hdl.handle.net/1721.1/6272~AIM-410
[47] G. Agha and R. Panwar, “An actor-based framework for heteroge-
neous computing systems,” in Proc. Workshop Heterogen. Process.,
Beverly Hills, CA, USA, 1992, pp. 35–42.
[48] H. G. Baker, Jr., “Actor-systems for real-time computation,” Ph.D. dis-
sertation, Lab. Comput. Sci., Mass. Inst. Technol., Cambridge, MA, Seddik Bacha (M’08) received the Engineer
USA, 1978. and Magister degrees from École Nationale
[49] National Instruments. (May 24, 2013). Actor Polytechnique de Algiers, Algeria, in 1982 and
Framework Template Documentation. [Online]. Available: 1990, respectively, and the Ph.D. degree from
http://www.ni.com/white-paper/14115/en the Institut National Polytechnique de Grenoble,
[50] Electric Power Research Institute. (2013). OpenDSS PVSystem France, in 1993. He joined the Laboratoire
Model.pdf (file), Supplied as Part of the OpenDSS Documentation. d’Electrotechnique de Grenoble in 1990 and in
[Online]. Available: http://sourceforge.net/projects/electricdss/ 1998, he was habilitated to conduct research. He
[51] R. F. Arritt and R. C. Dugan, “The IEEE 8500-node test feeder,” in is currently a Professor with the University Joseph
Proc. IEEE PES Transm. Distrib. Conf. Expo., New Orleans, LA, USA, Fourier of Grenoble and the SuperGrid Institute,
2010, pp. 1–6. Lyon, France.