Download as pdf or txt
Download as pdf or txt
You are on page 1of 10

IET Generation, Transmission & Distribution

Research Article

ISSN 1751-8687
Sectionalising methodology for parallel Received on 25th July 2014
Revised on 20th December 2014
system restoration based on graph theory Accepted on 14th January 2015
doi: 10.1049/iet-gtd.2014.0727
www.ietdl.org

Jairo Quirós-Tortós ✉, Mathaios Panteli, Peter Wall, Vladimir Terzija


School of Electrical and Electronic Engineering, The University of Manchester, Manchester, M13 9PL, UK
✉ E-mail: jairo.quirostortos@manchester.ac.uk

Abstract: Parallel power system restoration (PPSR) restores isolated sections (islands) of the network in parallel, thus the
overall restoration process is accelerated. These islands are defined during the preparation stage of PPSR as part of a
sectionalising strategy (SS). During this process, it is important that the operators only use updated post-blackout
system information. This study proposes a new methodology based on the ‘cut-set’ matrix defined in graph theory,
which can identify a shortlist of suitable SSs that satisfy the critical PPSR constraints in a few minutes. This short list of
SSs, not identified in previous works, can be presented to the operators to help them select a restoration plan that is
tailored to the specific changes in topology and asset availability that the blackout has caused. The methodology is
illustrated using the IEEE 9-bus system, and validated using the IEEE 118-bus and the Polish 3375-bus system to
demonstrate the efficiency of the new approach for large-scale networks. Multiple case studies are developed to
demonstrate the adaptability of the methodology to different system conditions, for example, the unavailability of
assets. In every case, the methodology quickly identified a number of SSs that create suitable islands for parallel
restoration.

1 Introduction A properly defined SS should limit any delay in the reconnection


of the critical load (CL), that is, load that must be rapidly restored [6],
The role of power system restoration is to quickly and safely restore and significant load, that is, the non-critical load, as capacity
systems that have experienced a partial or complete blackout [1]. becomes available [6] and it should also accelerate the
System’s blackstart resources are critical elements of any resynchronisation of the islands [7, 8]. The design of a suitable SS
restoration plan, as these units provide the cranking power to the that will help ensure a reliable PPSR must take into account the
non-blackstart units [2]. It is possible to use centralised cranking following constraints [6–11]:
sources to provide cranking power to the entire system as part of a
‘build-down’ strategy [1–4]; however, for a large power system 1. Each island must have at least one blackstart unit and the cranking
this can be excessively time consuming [1–6]. Most utilities do groups must be preserved.
not pursue this time consuming build-down process and instead 2. Each island should have sufficient capacity to maintain a
use parallel power system restoration (PPSR) to accelerate the satisfactory frequency by matching generation and load.
restoration process [1–11]. PPSR achieves this acceleration by 3. Each island should have sufficient voltage control resources to
sectionalising the area of the system to be restored into islands that maintain a suitable voltage profile.
contain sufficient cranking sources and, as such, can be restored in 4. The tie-lines between the islands should all have monitoring
parallel. equipment that allows the synchronisation of the adjacent islands
The parallel restoration of a power system should be executed in that they separate to be measured.
three stages [6]: (i) preparation, (ii) system restoration and (iii) 5. Each island should be monitored at the system control centre to
load restoration. Several tasks must be performed during each ensure correct operation.
stage, the most common of which are shown in Fig. 1.
This paper proposes a methodology for the design of The benefits of PPSR mean that most Transmission System
sectionalising strategies (SSs), which could support transmission Operators (TSOs) have developed SSs, for example, the Mexican
system operators (TSOs) during the preparation stage of PPSR (see system [9], the PJM interconnection [10] and the British network
Fig. 1) [6–11]. A SS defines a set of lines that will separate the [11]. To simplify the design process, these SSs have been
system into islands. A suitable SS creates islands that satisfy a determined based on factors like historical asset ownership and
number of critical constraints, which ensure that the islands operator experience. Although this practice has enabled TSOs to
created can be restored in parallel [6–8]. A properly planned effectively restore their system in the past [9–11], it provides no
PPSR, with a suitable SS, will accelerate the restoration of a guarantee of the quality of the restoration planning or the proper
system after a blackout. exploitation of the physical properties of the system, and may
The information used to design a suitable SS is collected after the have led to unnecessary delays in the past [7, 8]. Therefore the
blackout has occurred, as the actual system topology and the study of more quantitative methods for the proper design of a
availability of its elements must be considered [6–8]. The most SS has been an area of much interest.
important information includes [6–11]: the available blackstart Heuristic search algorithms are proposed in [7, 8], but these are
capability, the availability of interconnection assistance, the status limited to designing a SS that satisfies constraints 1 and 2. A
of non-blackstart units, the cranking groups (a cranking group method based on a genetic algorithm is introduced in [12] to
consists of at least one blackstart unit and a group of define a SS that creates islands with the minimum amount of load
non-blackstart units [4, 6]), the status of lines and circuit breakers unserved. Spectral graph clustering (the potential benefits of which
and the predicted load levels. to the restoration process and to splitting power networks are

IET Gener. Transm. Distrib., 2015, Vol. 9, Iss. 11, pp. 1216–1225
1216 & The Institution of Engineering and Technology 2015
Fig. 1 Three main stages of PPSR

studied in [13, 14], respectively) was recently used in [15, 16] to This paper is organised as follows. Section 2 presents the
design a suitable SS that considers the first three constraints. background of graph theory. Section 3 describes the proposed SS
Although these works are effective, the nature of their solutions methodology. Simulation results for the IEEE 9 and 118 bus test
limits them to providing only a single SS. However, any systems and the Polish 3375-bus case are presented in Section 4 to
circumstance that renders this single solution unsuitable would demonstrate the effectiveness of the methodology and its
prevent the proposed methods from supporting the operator in adaptability to different system conditions. Section 5 concludes the
defining a PPSR plan. Therefore a method that can define multiple paper.
similar SSs of comparable quality would likely be of more benefit
to system operators’ decision-making, as they will be able to select
the most suitable SS based upon their knowledge, experience and 2 Preliminaries on graph theory
the post-blackout conditions. The shortcomings of the
aforementioned practices highlight the need for more effective The basis of graph theory is creating a standardised mathematical
restoration planning techniques. representation of a complex system, for example, a power system,
This paper presents a SS methodology that uses the graph that can be viewed as a graph of nodes and edges [17]. A graph
theory-based concept of a constrained ‘cut-set’ matrix [17], also captures the topology of the network through the connectivity of
known as the ‘cut-edge’ incidence matrix [18], to create a set of the nodes, which is represented by the incidence matrix [17, 18],
suitable SSs that satisfy constraints 1–4. In this paper, these and functional information about the system (e.g. power balance)
constraints are called the critical PPSR constraints. It is assumed through weight factors, as detailed below.
that each island can be monitored by the control centre, which
satisfies constraint 5. The new methodology can determine a set of
suitable SSs within a few minutes, which is negligible when 2.1 Graph theory definitions
compared with the actual duration of restoration. These SSs are
designed using information collected after the blackout (cranking A power system affected by a complete blackout, which has n buses,
groups, the status of lines and circuit breakers and the predicted m generators, l branches and b blackstart units, can be represented as
load levels), which allows them to adapt to the changes in a connected undirected graph G = (V, E, ρ), where V = {v1, …, vn} is
topology and asset availability that a blackout will cause. the set of nodes (buses), E = {e1, …, el} is the set of edges
However, the methodology may identify several thousand suitable (branches), and ρ is a function that assigns weight factors to the
SSs for large power systems and presenting the TSO with such a nodes. An example of the graph representation of a system is
large number of SSs would not be a useful form of support. given in Fig. 2.
Therefore this set of suitable SSs is reduced to a shortlist of five For solving the SS-problem, the node subsets VGN , V and
(the selection of five SSs is arbitrary and could be any number the VB , VGN are defined to represent the m generation-buses
operator considers suitable) SSs that can be presented to the TSO (generation-nodes) and the b blackstart units of the system,
to support their restoration planning. respectively. Furthermore, the subset VLD = V\VGN (where \ denotes

Fig. 2 IEEE 9-bus system


a Single line diagram
b Simplified graph

IET Gener. Transm. Distrib., 2015, Vol. 9, Iss. 11, pp. 1216–1225
& The Institution of Engineering and Technology 2015 1217
the set-theoretic difference and defines VLD as the set of nodes in V
that do not appear in VGN) is defined to represent the n − m
load-buses (load-nodes). Finally, the subset VCLD , VLD represents
the load-buses that have CLs connected to them. Any load
connected to a generation-bus is considered to be a CL as these
commonly represent local generator services [10, 11].
The number ui = ρ(vi) is the weight factor associated with the node
vi ∈ V, and is defined as follows

ui = PGN,i
max pre
− ai PLD,i , for all,,vi [ V (1)

where Pmax pre


GN,i and PLD,i are the active power generation capability and
the predicted active power demand at bus i, respectively. The value ui
Fig. 3 IEEE 9-bus system sectionalised across the cut-set ES = {e2, e8}
provides a measure of the balance at each bus between the available
generation and the predicted load to be restored prior to the
resynchronisation of the islands. The scalar αi represents the as follows
proportion of the load at bus i that will be restored before 
c(Vr ) = ui (3)
resynchronising the islands.
vi [Vr
As the entire CL must be restored as quickly as possible, αi =
100% if vi ∈ VCLD. However, the amount of the significant load where ui is the node-weight given by (1).
that must be restored prior to resynchronisation of the islands can The value of c(Vr) is used to determine if the islands have a
vary from 40 to 75% of the total system load [10, 11]; therefore satisfactory load-generation balance (constraint 2). For example,
αi = 70% is arbitrarily selected here for all non-critical loads, the total weight of the subgraphs shown in Fig. 3 are c(V1) = u1 +
vi∉VCLD. In practical implementations, the definition of αi for the u4 + u6 = 187 MW and c(V2) = u2 + u3 + u5 + u7 + u8 + u9 = 345 MW,
nodes vi∉VCLD would require an accurate estimation of the load indicating a surplus of generation in both islands.
levels. A higher value of αi would reduce the number of suitable For a suitable SS, the weight of each subgraph must be positive.
SSs, as this value is used when calculating if there is sufficient This will ensure that the majority of the load can be restored and
generation capacity (Constraint 2) in each island to support the ease the reconnection of the islands, as islands with more load
anticipated loading level (see Section 3.3). restored are more likely to endure any voltage related issues
The graph representation used in this paper includes the locations during the resynchronisation of the islands [7–11, 19–21].
of the generators and loads, the terminal buses of a branch, the
maximum active power output of the generators and the predicted
active power load. Using this graph representation is more 2.4 Label of the node
convenient [17] and also takes advantage of the benefits of
implementing the well-established and powerful techniques from The label of the node defines the island that the node belongs to. In
graph theory, for example, the concept of a constrained ‘cut-set’ Fig. 3, the label of the nodes {v1, v4, v6} is ‘1’ and the label of the
matrix [18] that is used for determining the cut-sets in a graph. nodes {v2, v3, v5, v7, v8, v9} is ‘2’, as they belong to Island 1 and
Fig. 2 contrasts a system model and a graph representation of the 2, respectively.
IEEE 9-bus system. Fig. 2a shows the single line diagram and In the SS-problem, some nodes must have a specific label. For
Fig. 2b its graph representation. Fig. 2b also shows the node-weights. instance, the generation-nodes that represent the generators in the
same cranking group must have the same label to preserve the
integrity of these groups (Constraint 1). In contrast, some nodes
2.2 Incidence matrix (mainly load-nodes) are defined as ‘free-nodes’, as they can belong
to any of the islands. The definition of the island to which each
This paper adopts the definition of the incidence matrix M of an node belongs represents a combinatorial problem that increases in
undirected graph that is proposed in [17], which is defined as size with the size of the power system [7, 8]. Defining the
follows for an arbitrary orientation of the edges free-node labels creates multiple combinations that are used to
build the indicator matrix.

⎨1 if edge ek is incident from node vi
M := [mik ] = −1 if edge ek is incident to node vi (2) 2.5 Indicator matrix

0 otherwise The indicator matrix X contains each combination of node labels that
would partition a graph G into subgraphs. The combinations in X
Each column of the matrix M represents the incidence between the label certain nodes to be in a given island and combine the
nodes vi ∈ V and the edge ek ∈ E. free-node labels. This is a combinatorial problem that can be
effectively solved by computing the permutations with repetitions
[22] of the free-node labels. Each combination in X also includes
2.3 Cut-set and weight of the subgraphs the labels for the generation-nodes and any other constrained nodes
so that each combination supplies a full set of labels that defines
A ‘cut-set’ ES , E of a graph G consists of the edges that must be the island each node will be clustered into. Thus, the ij-entry of the
removed to split G into two disjoint subgraphs G1 and G2. For matrix X is equal to the number (label) r = 1, 2 assigned to the
example, ES = {e2, e8} is a cut-set of the graph shown in Fig. 2b. node vi ∈ V for the jth combination. Therefore for β ‘free-nodes’
Each subgraph created by a cut-set corresponds to one of the node there are ζ = 2β possible combinations, that is, the size of X is n × ζ.
sets V1 and V2, such that V1 ∪ V2 = V and V1 ∩ V2 = Ø (see Fig. 3). For example, consider the partition of the graph shown in Fig. 2b
A cut-set contains the edges that represent the lines to be included into two subgraphs G1 and G2. Nodes v1 and v4 are constrained to
in the SS and each node subset consists of the nodes that represent belong to Island 1 (node v1 is the cranking source in Island 1 and
the buses in each island. node v4 is connected to it by a transformer, and transformers
The weight of the subgraph for each node subset can be defined as cannot be included in SSs [6]). Similarly, nodes v2, v3, v7 and v9
the difference between the available power generation capacity and are constrained to be in Island 2. Finally, the ‘free-nodes’ are v5,
the predicted load level within the island represented by the node v6 and v8 (β = 3) hence eight (ζ = 23) combinations exist. Equation
subset. This value is denoted by c(Vr), r = 1, 2, and is calculated (4) shows the transpose of X – note that the size of X in (4) is 9 × 8.

IET Gener. Transm. Distrib., 2015, Vol. 9, Iss. 11, pp. 1216–1225
1218 & The Institution of Engineering and Technology 2015
Each combination in (4) defines a label (1 or 2) for each node, and The ik-entry of C is non-zero if the implementation of the ith cut-set
multiple nodes are constrained to have the same label for every requires the edge ek to be removed to partition G. The sign of the
combination (e.g. the label for v1 is 1 in every combination). If a ik-entry represents the alignment of the edge with the cut-set (see
node has the same label in every combination, then this node will [17] for more details). Each cut-set included in the matrix C
definitely be part of the corresponding island. This allows the represents the edges that must be removed to separate the elements
cranking groups to be preserved in every combination. This of G that have different labels in the corresponding combination
contrasts to the behaviour of the free-node labels, which vary shown in X. It should be recognised that, while the cut-sets in C
depending on the combination, as each combination in X will ensure that no nodes are connected to a node with a different
represents a different combination with repetitions of the free-node label, these cut-sets will not guarantee that G will be separated
labels {1, 2} [22] into two graphs, see Fig. 4 for examples of this. The matrix C for
⎡ ⎤T the graph shown in Fig. 2b, considering the free-nodes v5, v6 and
v1 v2 v3 v4 v5 v6 v7 v8 v9 v8, is
⎢ ⎥
⎢ comb1 1 2 2 1 1 1 2 1 2 ⎥ ⎡ ⎤
⎢ ⎥
⎢ comb2 1 2 2 1 1 1 2 2 2 ⎥ e1 e2 e3 e4 e5 e6 e7 e8 e9
⎢ ⎥ ⎢ ⎥
⎢ comb3 1 2 2 1 1 2 2 1 2 ⎥ ⎢ cut-set1 0 ⎥
⎢ ⎥ ⎢ 0 0 1 0 1 1 0 1 ⎥
X = ⎢ comb 1 2 2 1 1 2 2 2 2⎥ (4) ⎢ cut-set2
⎢ 4 ⎥ ⎢ 0 0 1 0 0 0 0 1 0 ⎥⎥
⎢ comb5 1 2 2 1 2 1 2 1 2 ⎥ ⎢ cut-set3 −1 ⎥
⎢ ⎥ ⎢ 0 0 1 0 1 1 0 0 ⎥
⎢ comb6 1 2 2 1 2 1 2 2 2 ⎥ C = ⎢ cut-set4 −1 ⎥ (7)
⎢ ⎥ ⎢ 0 0 1 0 0 0 0 0 ⎥
⎣ comb7 1 2 2 1 2 2 2 1 2 ⎦ ⎢ cut-set5 0 −1 0 0 1 1 0 1 0 ⎥
⎢ ⎥
comb8 1 2 2 1 2 2 2 2 2 ⎢ cut-set6 0 −1 0 0 0 0 0 1 0 ⎥
⎢ ⎥
⎣ cut-set7 0 −1 0 0 1 1 0 0 −1 ⎦
2.6 Constrained cut-set matrix cut-set8 0 −1 0 0 0 0 0 0 −1

A cut-set matrix, also known as a ‘cut-edge’ incidence matrix, is a


For the given constraints on the node labels, there are eight (ζ = 23)
matrix used in graph theory to represent all of the possible cut-sets
cut-sets that separate the nodes with different labels and these may
in a graph [17]. The constrained cut-set matrix C is an extension
not satisfy the PPSR constraints.
of the cut-set matrix that enables us to constrain certain nodes to
The nature of the combinatorial optimisation problem and the
be included into a specific group. The constrained cut-set matrix C
topology of the power system may produce a constrained cut-set
of the graph G is the ζ × l matrix that is defined as follows [17]
matrix C that contains cut-sets that are not suitable for
sectionalising the system in blackout. An unsuitable cut-set may
1 if ith cut-set contains edge ek [ E
C := [c]ik = (5) inadvertently sectionalise the system in blackout into more than
0 otherwise two islands, and some of these may not have cranking sources.
For example, the cut-sets {cut-set1, cut-set3, cut-set5, cut-set7} in
The matrix X contains the labels of the nodes obtained from the (7) create the islands shown in Fig. 4.
combinatorial optimisation and the incidence matrix M contains These cut-sets produce four islands, where the bus represented by
the arbitrary orientation of the edges that connect the nodes v8 is isolated from the system, and two of the islands created lack
labelled in X. We have found that these can be used to compute cranking resources, which are represented by v1 and v2. Hence,
the cut-set matrix C, for the case of two islands, as follows these cut-sets are deemed to be unsuitable.
To determine the suitable cut-sets, the rank of the graph created by
C = X TM (6) the cut-set must be computed. The rank of a graph is by definition

Fig. 4 Unsuitable cut-sets for the IEEE 9-bus test system

IET Gener. Transm. Distrib., 2015, Vol. 9, Iss. 11, pp. 1216–1225
& The Institution of Engineering and Technology 2015 1219
equal to the difference between the number of nodes, n, and the assigning each non-blackstart unit to one of these groups of
number of connected components in the corresponding graph [17]. blackstart units. However, the definition of these cranking groups
For instance, the rank of the graph shown in Fig. 2b is eight, as is a complex process that requires detailed and comprehensive
the number of connected components is one, whereas the rank of system studies [2–4], which are beyond the scope of this paper.
each graph shown in Fig. 4 is five, as four connected components Therefore the cranking groups used in this paper are those found
exist. The rank of a graph is also equal to the rank of the using the method presented in [23], which optimally groups the
Incidence matrix M [17]. Therefore to ensure that only two islands machines in such a way as to minimise the duration of the
are created by a cut-set, the rank of the graph produced by that cranking task.
cut-set must be equal to n − 2. After determining the rank of the Note, if necessary, any reserve generator can also be constrained
graphs produced by each cut-set, the unsuitable cut-sets can be to belong to a given island. To do this, the label of the
excluded from the set of candidate solutions in C. By doing this, generation-node representing this bus must be constrained in X to
the cut-sets that produce more than two islands are excluded from belong to the specific island – as in the case of the cranking groups.
consideration as suitable SSs. The set of suitable cut-sets in (7) is
{cut-set2, cut-set4, cut-set6, cut-set8}. Step 1: Label in X the generation-nodes vi ∈ VGNr – the nodes within
the same cranking group – with the same number in every
combination to ensure that they are separated into different islands.
3 New methodology for designing suitable This ensures that there is at least one blackstart unit within each
sectionalising strategies island and that the cranking groups are preserved. For instance,
note in (4) that the label of the cranking groups VGN1 = {v1} and
This section describes the proposed methodology for determining VGN2 = {v2, v3} are the same.
several suitable SSs that satisfy the critical PPSR constraints,
based on the concept of a constrained cut-set matrix. Without loss
of generality, the step-by-step execution of the proposed 3.2 Step 2: Exclusion of lines
methodology for the case of two islands is presented below. The
new methodology can produce more islands if required, as detailed To satisfy Constraints 3 and 4, it is necessary to exclude certain
in Section 3.3. Fig. 5 shows the flowchart of the proposed branches from the solution space. As discussed in Section 1, a
methodology. As it can be observed, three main steps are needed: suitable SS cannot include lines that maintain the voltage stability
(i) Ensuring blackstart availability and preserving the cranking within the islands or lines that lack the ability to monitor
groups; (ii) Exclusion of lines; and (iii) Determination of SSs. synchronisation.

Step 2(a): To define the critical lines that maintain system voltage
3.1 Step 1: Ensuring blackstart availability and stability (Constraint 3), it is common practice in the UK to
preserving the cranking groups consider that the tie-lines between the islands are open during the
restoration process [11]. The critical lines are classed as those lines
If an island does not contain at least one blackstart unit (Constraint 1)
that are the most severe outages identified by the N − 1
it cannot be restored separately, as there will be no cranking power
contingency analysis method available in DIgSILENT Power
available to energise the non-blackstart units. Blackstart units
Factory [24]. This means that a line is critical if its disconnection
typically only supply cranking power to the non-blackstart units
produces an excessive reactive power difference (i.e. a
that are within the same cranking group [4]. The location and
significantly higher difference than that which is seen for other
availability of the blackstart units are commonly known after the
disconnections) between the pre-contingency and the
assessment of the system in blackout [1–4, 6], and this
post-contingency, as it is defined in current practices of many
information is used to define the subsets VB1 and VB2. The
TSOs, such as PJM (USA) and National Grid (UK) [10, 11].
cranking groups, denoted by VGN1 and VGN2, are then formed by
These lines are represented by the subset EVS , E.
Step 2(b): The SS defines the tie-lines that the breakers of which will
be closed to resynchronise the islands; therefore the operator must
possess the ability to monitor the phase angle across the lines that
separate the islands to facilitate a successful resynchronisation
(Constraint 4) [6]. Hence, lines without monitoring equipment are
also excluded from the solution space. As with the lines critical to
voltage stability, a subset EC , E is used to represent the branches
that cannot measure resynchronisation, for example, they are not
equipped with synchro-check relays [1].
Step 2(c): When the subset of edges for each type of branch to be
excluded is found they can be combined into a single subset of
excluded edges, that is, EEX = EVS ∪ EC. The edges ek ∈ EEX are
then excluded from the solution space by labelling the nodes vi
and vj incident to the edge ek ∈ EEX with the same number r = 1,
2. This mechanism for excluding edges from the SSs is not limited
to the branch types considered in this paper. In practice, it could
be extended to any branch, or branch type, that the TSOs deemed
unsuitable for use as a tie-line between islands. Examples of
unsuitable branches are given in [2, 3, 6–8].

3.3 Step 3: Determination of sectionalising strategies

Step 3(a): Calculate the arbitrary Incidence matrix M using (2).


Step 3(b): Create the indicator matrix X. It contains all the possible
combinations of node labels. For this, only the ‘free-nodes’ are used
in the combinatorial optimisation (i.e. permutations with repetitions
Fig. 5 Flowchart of the proposed methodology [22]). This creates a number of combinations of node labels. Fixing

IET Gener. Transm. Distrib., 2015, Vol. 9, Iss. 11, pp. 1216–1225
1220 & The Institution of Engineering and Technology 2015
the node labels of the generators and excluded lines reduces the Table 1 Generation-load difference within the islands for the IEEE 9-bus
number of ‘free-nodes’. This also limits the size of the matrix X, test system
which reduces the number of SSs that must be considered, Sectionalising strategy Generation-load difference, MW
accelerating thus the process of identifying a shortlist of suitable SSs.
Step 3(c): Compute using (6) the constrained cut-set matrix C. Island 1 Island 2
Step 3(d): Determine the rank of the graph produced by each cut-set
included in C to evaluate the suitability of each cut-set. To define a 5–7; 6–9 62 470
cut-set as suitable, the rank of the induced graph (split into 5–7; 4–6 125 407
4–5; 6–9 187 345
subgraphs) must be equal to (n − 2). This creates a set of cut-sets 4–5; 4–6 250 282
that produce two islands that satisfy Constraints 1, 3 and 4.
Step 3(e): Create islands with sufficient generation capacity
(Constraint 2). To do so, compute the n-vector u, which is a
column vector with the ith row equal to the node weight ui given system with the branch and the system without the branch. For
by (1). For instance, the vector u for the IEEE 9-bus system is this particular system, the branches that produced excessive
created using the information shown in Fig. 2b. The vector u is reactive power difference correspond to the transformers that
then used to calculate the weight of the subgraphs c(Vr) according connect the generators with the rest of the system. Therefore the
to (3). Aforementioned, c(Vr) represents the difference between set of critical lines is EVS = {e1, e4, e7}.
the power generation capacity and the predicted load level within In Step 2(b) it is assumed that every line can measure
each island. synchronisation with adjacent islands. Thus, the set EC = Ø is
Step 3(f): Define the final set of suitable SSs, which could contain defined. Therefore in Step 2(c) the set of excluded edges is simply
multiple cut-sets, as those cut-sets that sectionalise the system into the set of critical lines found in Step 2(a), that is, EEX = EVS = {e1,
two islands with positive subgraph weight in both islands. e4, e7}. These results are used to build the indicator matrix X, as
Step 3(g): Reduce the number of cut-sets to significantly benefit the in (4).
system operators’ decision-making. This is needed because the In Step 3, the methodology creates the arbitrary incidence matrix
number of suitable cut-sets previously determined could be large. M (2) of the graph, and then computes the indicator matrix X, shown
The power imbalance within the islands and the number of in (4). It should be noted that the label of the node v1 must be ‘1’, as
tie-lines (lines to be reconnected to resynchronise the islands) for the generator at bus 1 must be in one island (cranking group VGN1),
each cut-set are used to identify a shortlist of solutions that can be and the label of the nodes v2 and v3 must be ‘2’, as the generators at
presented to the operator to effectively support their buses 2 and 3 must be in the other island (cranking group VGN2).
decision-making. The shortlist will contain the five cut-sets with Furthermore, the label of the node v4 must be ‘1’, and the label of
sufficient generation capacity and the minimum number of the nodes v7 and v9 must be ‘2’. This is because these nodes are
tie-lines, as detailed in Section 4.2. When considering the length connected to the generation-nodes by an excluded edge,
of the shortlist it is important to balance the desire to maximise specifically an edge representing a transformer. The ‘free-nodes’
the range of options offered to the operator against the danger of are v5, v6 and v8.
overwhelming them with numerous, similar solutions and thereby It is important to remember that the labels given to the free-nodes
delaying the planning process. It is important to emphasise that indicate that they can be in any island. The methodology then uses
this shortlist of solutions will all satisfy the critical PPSR the matrices X and M to compute the constrained cut-set matrix C
constraints described above. that is shown in (7). It then evaluates whether the cut-sets included
Step 3(h): When more than two islands are required, the set V1 or V2 in C are suitable or not. In this case, there are four suitable
that contains more than one blackstart unit should be selected as the cut-sets. Although the node v8 was defined as a ‘free-node’, it
node set of a new graph. The methodology should then be cannot appear in Island 1 because of the inherent topology of the
implemented again. This process should be repeated until the system. It is important to mention that the methodology is capable
required number of islands is achieved. of detecting these spurious solutions and excluding them from the
set of suitable SSs.
The methodology then computes the difference between the
4 Simulation results power generation capacity and the predicted system load within
each possible island for the four suitable cut-sets. The results are
The new methodology is now tested using the IEEE 9- and 118-bus shown in Table 1 and the third SS was arbitrarily selected to be
test systems, as well as the Polish 3375-bus case. All simulations are shown in Fig. 6.
performed in MATLAB [25], using the values of PGN,i max pre
and PLD,i In this case, the four cut-sets result in islands with positive
available in MATPOWER [26]. All times quoted are based upon generation-load difference, that is, they are all suitable. When
simulations performed on a PC with a 2.93 GHz dual core CPU moving nodes between the islands, it should be noted that an
and 4 GB RAM. action that causes an increase in the generation-load difference in
Island 1 will cause a decrease in the generation-load difference in

4.1 IEEE 9-bus test system

The methodology is applied to the IEEE 9-bus system shown in


Fig. 2. It is assumed that every blackstart unit, non-blackstart unit,
line and breaker is available after the blackout. If any of these
elements were to be unavailable, it would result in a reduction in
the number of suitable cut-sets, as shown in Section 4.2.
The following subsets VB = {v1, v2} and VCLD = {v5, v8} are
considered (more details have been presented above). Therefore in
Step 1, the cranking groups are assumed to be VGN1 = {v1} and
VGN2 = {v2, v3} and their labels are fixed in the X matrix
accordingly (see (4) for more details).
In Step 2(a) the critical lines that maintain voltage stability are
identified. These lines are determined using the N − 1 contingency
analysis method available in DIgSILENT Power Factory [24]. The
pre-blackout system is used for this analysis and every branch is
disconnected to evaluate the reactive power change between the Fig. 6 IEEE 9-bus test system sectionalised into two islands

IET Gener. Transm. Distrib., 2015, Vol. 9, Iss. 11, pp. 1216–1225
& The Institution of Engineering and Technology 2015 1221
Island 2. It is important to remember that the four suitable SSs satisfy Table 2 Generation-load difference within the islands in case study 1
the critical PPSR constraints. This set of SSs was computed in 10 ms, for the IEEE 118-bus test system
which is negligible when compared with the actual duration of PPSR Sectionalising strategy Number of Generation-load
[5]. tie-lines difference, MW

Island 1 Island 2
4.2 IEEE 118-bus test system
23–24;30–38;19–20;18–19;15–19; 6 804 2112
The IEEE 118-bus system shown in Fig. 7 is now used to validate the 15–33
methodology for a larger network. The assumed CLs are highlighted 13–15;14–15;15–17;15–19;23–24; 7 862 2054
in Fig. 7. It is considered that the blackstart units are located at buses 19–34;30–38
13–15;14–15;15–17;18–19;19–20; 7 894 2022
31 and 87. Two cases are examined here to show the adaptability 23–24;30–38
of the methodology to different system conditions and to study 13–15;14–15;15–17;18–19;19–20; 8 876 2040
the impact of the unavailability of system assets. Case study 1 30–38;24–70;71–72
assumes that every system element is available after the blackout 13–15;14–15;15–17;18–19;19–20; 8 884 2032
30–38;24–70;24–72
and case study 2 examines the unavailability of system assets
(a generator and a line) after the complete blackout.

4.2.1 Case study 1: Availability of all system assets: The


first case study assumes that every blackstart unit, non-blackstart experience, the most suitable for their system. These SSs were
unit, line and breaker is available after the blackout. For the found in 115 s, which is sufficiently fast for the purposes of
given blackstart units, the cranking groups are: VGN1 = {v10, v12, PPSR. The minimum and maximum generation-load difference
v25, v26, v31} and VGN2 = {v46, v49, v54, v59, v61, v65, v66, v69, v80, within Island 1 (Island 2) are 433 MW (2023 MW) and 893 MW
v87, v89, v100, v103, v111}. The critical lines that maintain system (2483 MW), respectively. Fig. 8 shows the distribution of the 324
voltage stability are found to be {5–8, 8–9, 9–10, 37–38, 38–65}. suitable cut-sets into five groups of 92 MW ranges (i.e. maximum
It is then assumed that every line can monitor synchronisation minus minimum generation-load difference over five).
with adjacent islands. The edges that represent transformers Fig. 8 shows that as the generation-load difference within Island 1
are excluded from the solution space by including the appropriate increases it decreases within Island 2. Given that the generation-load
edges in the subset EC. Finally, the subset of excluded-edges difference for Island 2 is significantly larger than that of Island 1 for
EEX = EVS ∪ EC is defined; these will be excluded from the SSs all of the suitable SSs, it seems prudent to select a cut-set from those
by labelling the applicable nodes with the same node label. that offer the largest generation-load difference in Island 1. This
The implementation of Step 3 then identifies 324 suitable SSs. reduces the list of possible SSs to 44 candidates.
This will be reduced to a final shortlist of five cut-sets as will be To further reduce the number of suitable SSs, the five cut-sets
explained below (Table 2). Generating a shortlist of suitable SSs within these 44 candidates with the minimum number of tie-lines
enables TSOs to select, based upon their knowledge and are presented as the final shortlist (see Table 2). Fig. 7 shows the

Fig. 7 IEEE 118-bus system sectionalised into two islands

IET Gener. Transm. Distrib., 2015, Vol. 9, Iss. 11, pp. 1216–1225
1222 & The Institution of Engineering and Technology 2015
Fig. 8 Distribution of the suitable cut-sets in case study 1 for the IEEE 118-bus system

SS from the first row of Table 2. A number of lines appear in several to the loss of this unit, the second scenario (Scenario B) considers
of these cut-sets, which indicates that certain lines are more likely to that the monitoring equipment on a line (15)–(19) is unavailable,
be used as tie-lines during restoration. This information could serve which is equivalent to losing a line in this context (Constraint 4).
to guide operators towards improving the reliability of these lines or In Scenario A the cranking groups are as defined previously,
developing specific plans to try and ensure that they will be available excluding the removal of the generator at bus 10, and the
during restoration. methodology computed 294 suitable cut-sets, which satisfy the
critical PPSR constraints in <2 min.
4.2.2 Case study 2: Unavailability of system assets: The Fig. 9 shows the distribution of these 294 suitable cut-sets (upper
second case study is divided into two scenarios that aim to row of x axis labels) into five groups of 68 MW ranges. It should
demonstrate the adaptability of the new methodology to different be noted that the unavailability of the generator at bus 10 results in
system conditions. It is initially assumed that the non-blackstart unit the reduction of the minimum and maximum generation-load
at bus 10 is unavailable after the blackout (Scenario A). In addition difference within Island 1 (Island 2) to 5 MW (2021 MW) and

Fig. 9 Distribution of the suitable cut-sets in case study 2 for the IEEE 118-bus system
Upper row of x-axis labels refers to the number of cut-sets in each range for Scenario A and the lower row of labels refers to Scenario B

IET Gener. Transm. Distrib., 2015, Vol. 9, Iss. 11, pp. 1216–1225
& The Institution of Engineering and Technology 2015 1223
345 MW (2361 MW), respectively. In practical implementations, the Table 4 Generation-load difference within the islands – Polish system
unavailability of a non-blackstart unit may require an update of the
Number of tie-lines Generation-load difference, MW
predicted load levels [3, 4].
As before, it would seem sensible to reduce the list of cut-sets to Island 1 Island 2
those that offer the largest difference for Island 1, as it has a much
smaller difference than that of Island 2 for every cut-set. Therefore 17 4600 13 710
the five cut-sets with the minimum number of tie-lines from within 19 4570 13 740
the 33 suitable cut-sets with the largest imbalance in Island 1 are 19 4560 13 750
21 4540 13 770
used to form the final shortlist. 22 4530 13 780
In Scenario B, in addition to the unavailability of the
non-blackstart unit at bus 10, it is assumed that line 15–19 loses
monitoring capacity. Hence, it must not be included in the SS
(Constraint 4). The implementation of the methodology in this Table 5 Results of implementing the method available in [16] for the
Polish system
new scenario produces 214 suitable SSs. The distribution of these
is also represented in Fig. 9 (lower row of x axis labels). Note that Number of tie-lines Generation-load difference, MW
the ranges remain the same, as the unavailability of a line would
not reduce the available generation capacity. Finally, the Island 1 Island 2
unavailability of these two system assets reduces the number of
cut-sets that are in the range with the largest generation capability 25 4515 13 795
for Island 1 to 30. These results highlight the adaptability of the
methodology to unexpected topology changes following a
complete blackout. Finally, the shortlist of suitable cut-sets that are 4.3.1 Comparison with a spectral clustering-based
presented to the operators is reduced to five (see Table 3) for the method: This section compares the results of applying the
conditions (loss of two system assets) described above. proposed methodology and a spectral clustering-based method [16]
to the Polish system when all assets are available. Table 5 shows
the results of implementing the method presented in [16]. Note
4.3 Polish system
that this solution is in the range with the largest generation
The proposed methodology is now implemented for the Polish capability for Island 1. However, the number of tie-lines obtained
system (winter 2007–2008 evening peak) available in with the method in [16] is higher than the top five previously
MATPOWER [26]. This system consists of 3375 buses, 596 presented (Table 4). The implementation of the method [16]
generators and 4161 branches. This is used to demonstrate the required about 3 min to determine this single solution (whereas the
capability of the approach to determine multiple SSs for a real methodology presented in this paper required <8 min to find the
network in a short period of time, when compared with the actual top five from the set of 5000 suitable cut-sets). This highlights that
duration of restoration. The methodology finds more than 5000 the proposed methodology is very efficient determining multiple
suitable cut-sets for the creation of two islands in <8 min when SSs that can be used by system operators.
applied to this system and the number of tie-lines required by It must also be highlighted that the method in [16] can only
these suitable cut-sets ranges from 16 to 59. This shows the determine one single SS. However, it is known that system
suitability of the approach for determining cut-sets that can operators may select the actual SS based upon their knowledge
support operators’ decision-making process in a reasonable amount and experience. Hence, it would be more beneficial to provide
of time for even a large power system. Furthermore, the large them with a short list (e.g. up to five) of suitable SSs.
number of cut-sets found demonstrates that it is a necessity to
reduce the list of suitable cut-sets to a shortlist that can be
presented to the operator. 5 Conclusions
These cut-sets are organised in 50 groups of 90 MW, following
the same approach used for the IEEE 118-bus system. The This paper proposes a methodology that uses the concept of a
number of cut-sets that are in the range with the largest constrained cut-set matrix to determine sectionalising strategies for
generation capability (Island 1: 4510–4600 MW; Island 2: 13 710– PPSR, which enables operators to accelerate restoration. In
13 800 MW) for Island 1 is 125. The five cut-sets with the fewest contrast to other approaches, the methodology can determine a
tie-lines within these 125 cut-sets produce the final shortlist of short list of suitable strategies to sectionalise the blackout area,
suitable cut-sets (see Table 4) that are ultimately presented to while satisfying the constraint of blackstart availability within each
operators for the selection of the most suitable based on their island (i.e. preserving the integrity of the defined cranking groups),
knowledge and experience. creating islands with more generation capacity than the predicted
load level and excluding from the solution critical lines that
maintain system voltage stability and those that cannot measure
Table 3 Generation-load difference within the islands in case study 2 resynchronisation with adjacent islands.
(Scenario B) – IEEE 118-bus system
Satisfying these constraints provides sufficient resources to
Sectionalising strategy Number of Generation-load maintain the steady-state stability of the islands within acceptable
tie-lines difference, MW limits during the actual restoration. The proposed methodology is
intended for use during the preparation stage of the restoration
Island 1 Island 2
process, after updated post-blackout system information is
13–15;14–15;15–17;18–19;19–20;23– 7 344 2022
provided. The proposed methodology can explore the vast
24;30–38 combinatorial space to find multiple sectionalising strategies that
13–15;14–15;15–17;18–19;19–20;30– 8 334 2032 satisfy the given critical constraints. To facilitate system operators’
38;24–70;24–72 decision-making, nonetheless, the methodology produces a
13–15;14–15;15–17;18–19;19–20;30– 8 326 2040
38;24–70;71–72
shortlist of suitable sectionalising strategies. This shortlist can
13–15;14–15;15–17;18–19;19–20;23– 12 285 2081 ultimately be used by the operators to select the most suitable
24;15–33; 35–36;34–37;37–40;39– strategy based upon their knowledge and experience. This will
40;38–65 enable them to simultaneously restore more system components, as
13–15;14–15;15–17;18–19;19–20;23– 12 283 2083
24;15–33; 34–36;34–37;37–39;37–
each island is restored independently.
40;38–65 The proposed methodology has been tested and validated using
the IEEE 9-bus test system, the IEEE 118-bus test system and the

IET Gener. Transm. Distrib., 2015, Vol. 9, Iss. 11, pp. 1216–1225
1224 & The Institution of Engineering and Technology 2015
Polish 3375-bus system. Two cases for the 118-bus system have 7 Nezam Sarmadi, S.A., Dobakhshari, A.S., Azizi, S., Ranjbar, A.M.: ‘A
been examined to demonstrate the adaptability of the approach to sectionalizing method in power system restoration based on wams’, IEEE Trans.
Smart Grid, 2011, 2, (1), pp. 178–185
different system conditions, for example, unavailability of assets. It 8 Lin, Z.Z., Wen, F.S., Chung, C.Y., Wong, K.P., Zhou, H.: ‘Division algorithm and
was found that the loss of assets reduced the number of cut-sets interconnection strategy of restoration subsystems based on complex network
but multiple suitable sectionalising strategies were still obtained in theory’, IET. Gener. Transm. Distrib., 2011, 5, (6), pp. 674–683
less than two minutes. These results highlighted that the proposed 9 Gutierrez, J., Staropolsky, M., Garcia, A.: ‘Policies for restoration of a power
methodology is effective during the preparation stage of parallel system’, IEEE Trans. Power Syst., 1987, 2, (2), pp. 436–442
10 PJM, Pjm Manual 36: System Restoration’, (2013)
restoration, in terms of the number of suitable cut-sets presented to 11 National Grid Electricity Transmission, The Grid Code’, (2013)
the system operator. The Polish system was then used to compare 12 Afrakhteh, H., Haghifam, M.R.: ‘Optimal islands determination in power system
the effectiveness and benefits of the proposed methodology against restoration’, Iran. J. Sci. Technol., 2009, 33, (B6), pp. 463–476
a spectral clustering based method available in the literature. The 13 Edstrom, F., Soder, L.: ‘On spectral graph theory in power system restoration’.
new methodology found a shortlist of five cut-sets in eight Proc. ISGT Europe, Manchester, UK, December 2011, pp. 1–8
14 Quiros-Tortos, J., Sánchez-García, R., Brodzki, J., Bialek, J., Terzija, V.:
minutes for the Polish 3375-bus system, whereas the implemented ‘Constrained spectral clustering based methodology for intentional controlled
spectral clustering required 3 min to find a single solution. This islanding of large-scale power systems’, IET Gener. Transm. Distrib., 2015, 9,
highlighted that the methodology is computationally efficient (1), pp. 31–42
enough to be employed during the preparation stage of PPSR. 15 Quiros-Tortos, J., Terzija, V.: ‘A graph theory based new approach for
power system restoration’. Proc. IEEE PowerTech, Grenoble, France, June 2013,
pp. 1–6
16 Quirós-Tortós, J., Wall, P., Ding, L., Terzija, V.: ‘Determination of sectionalising
6 Acknowledgments strategies for parallel power system restoration: a spectral clustering-based
methodology’, Electr. Power Syst. Res., 2014, 116, pp. 381–390
The authors would like to thank the Engineering and Physical 17 Chen, W.K.: ‘Graph theory and its engineering applications’ (World Scientific
Publishing, 1997)
Science Research Council (EPSRC) grants EP/E009735/1 in the
18 Zhu, J.: ‘Power system applications of graph theory’ (Nova Science Publishers,
U. K, and The University of Costa Rica for their financial support. Inc., 2009)
19 Adibi, M.M., Milanicz, D.P., Volkmann, T.L.: ‘Asymmetry issues in power system
restoration’, IEEE Trans. Power Syst., 1999, 14, (3), pp. 1085–1091
7 References 20 Adibi, M.M., Kafka, R.J.: ‘Power system restoration issues’, IEEE Comp. Appl.
Power, 1991, 4, (2), pp. 19–24
1 Adibi, M.M., Clelland, P., Fink, L.H., et al.: ‘Power system restoration - a task 21 Adibi, M.M., Alexander, R.W., Avramovk, B.: ‘Overvoltage control during
force report’, IEEE Trans. Power Syst., 1987, 2, (2), pp. 271–277 restoration’, IEEE Trans. Power Syst., 1992, 7, (4), pp. 1464–1470
2 Adibi, M.M., Fink, L.H., Andrews, C.J., et al.: ‘Special considerations in power 22 Moler, C.: ‘Numerical computing with Matlab’ (Society for Industrial and Applied
system restoration’, IEEE Trans. Power Syst., 1992, 7, (4), pp. 1419–1427 Mathematics, 2004)
3 Adibi, M.M., Fink, L.H.: ‘Power system restoration planning’, IEEE Trans. Power 23 Sun, W., Liu, C.C., Zhang, L.: ‘Optimal generator start-up strategy for bulk power
Syst., 1994, 9, (1), pp. 22–28 system restoration’, IEEE Trans. Power Syst., 2011, 26, (3), pp. 1357–1366
4 Adibi, M.M., Milanicz, D.P., Volkmann, T.L.: ‘Remote cranking of steam electric 24 DIgSILENT PowerFactory
stations’, IEEE Trans. Power Syst., 1996, 11, (3), pp. 1613–1618 25 MATLAB and Statistics Toolbox Release 2012b, (The MathWorks Inc., Natick,
5 Adibi, M.M., Milanicz, D.P.: ‘Estimating restoration duration’, IEEE Trans. Power Massachusetts, United States, 2010)
Syst., 1999, 14, (4), pp. 1493–1498 26 Zimmerman, R.D., Murillo-Sánchez, C.E., Thomas, R.J.: ‘Matpower: steady-state
6 Fink, L.H., Liou, K.L., Liu, C.C.: ‘From generic restoration actions to specific operations, planning, and analysis tools for power systems research and education’,
restoration strategies’, IEEE Trans. Power Syst., 1995, 10, (2), pp. 745–751 IEEE Trans. Power Syst., 2011, 26, (1), pp. 12–19

IET Gener. Transm. Distrib., 2015, Vol. 9, Iss. 11, pp. 1216–1225
& The Institution of Engineering and Technology 2015 1225

You might also like