Access the full text.

Sign up today, get DeepDyve free for 14 days.

Electrical Engineering and Systems Science
, Volume 2019 (1905) – May 29, 2019

/lp/arxiv-cornell-university/multi-layer-spectral-clustering-approach-to-intentional-islanding-in-katyD2aMkA

- ISSN
- 2196-5625
- eISSN
- ARCH-3348
- DOI
- 10.1007/s40565-019-0554-1
- Publisher site
- See Article on Publisher Site

Multi-Layer Spectral Clustering Approach to Intentional Islanding In Bulk Power Systems Faycal Znidi, Hamzeh Davarikia, Kamran Iqbal, Masoud Barati using combinatorial optimization approaches. The Abstract Intentional controlled islanding (ICI) is a final inclusion of reactive power or voltage in the constraints resort for preventing a cascading failure and catastrophic leads to a mixed integer nonlinear program (MINLP), power system blackouts. This paper proposes a controlled which is, in general, difficult to be solved than nonlinear islanding algorithm that uses spectral clustering over multi- programming problem (NPP) and mixed integer linear layer graphs to find a suitable islanding solution. The multi- programming (MILP) problem [1, 2]. Therefore, linear DC criteria objective function used in this controlled islanding power flow has often been used in literature resulting in a algorithm involves the correlation coefficients between bus MILP problem that promises a better computational burden. frequency components and minimal active and reactive Additionally, some other methods consider only the active power flow disruption. Similar to the previous studies, the power in system partitioning. As an example, in [2], it is algorithm is applied in two stages. In the first stage, groups claimed that local reactive power compensators can be used of coherent buses are identified with the help of modularity to compensate reactive power imbalance. In [3], a MILP- clustering using correlation coefficients between bus based splitting strategy is proposed to manage energy frequency components. In the second stage, the ICI solution production and demand. In this methodology the reactive power is viewed as a local issue and can be handled with with minimum active and reactive power flow disruption local reactive power compensators and only active power is and satisfying bus coherency is determined by grouping all considered in the splitting scheme. However, the reactive nodes using spectral clustering on the multi-layer graph. power plays a significant role in supporting the voltage Simulation studies on the IEEE 39-bus test system profile, and a significant mismatch of the reactive power demonstrate the effectiveness of the method in determining supply and demand causes high or low voltage conditions an islanding solution in real time while addressing the within islands. In [4], the MILP-based optimization method generator coherency problem. for controlled islanding disregards the generator coherency constraint, which is one of the most important requirements Keywords Constrained spectral clustering, Controlled in the islanding solution. The optimization-based islanding islanding, Bus coherency, Multi-layer graphs, Normalized algorithms are proposed in [5, 6], aiming to find the spectral clustering boundaries of electric islands. Utilizing the mathematical programming for islanding solution requires different set of constraints to ensure the islands integrity and feasibility, 1 Introduction including not limited to power balanced, connectivity, and operational constraints. In the other hand, the graph-based Intentional controlled islanding (ICI) has been islanding solutions automatically satisfies the connectivity proposed as a corrective measure of last resort to split the constraints, since the solution is sought trough the power system into several sustainable islands and prevent minimum cuts in the graph. cascading outages. Most approaches to islanding aim to In [7], the constrained spectral clustering is used to find find, as a primary objective, electromechanically stable islanding boundary with minimal power flow disruption. In islands with minimal load shedding. To find a reasonably [8], a binary particle swarm optimization (BPSO) seeking good islanding solution, all subsystems must satisfy some Pareto non-dominated solutions algorithm is presented to constraints, such as power flow disruption, generator find islands containing coherent generator groups with coherency, transient stability, etc. [1]. minimal power imbalances. BPSO being a stochastic Traditionally, the islanding problem has been solved evolutionary algorithm, multiple runs of the algorithm are needed to determine if the results are consistent. In [9], a where each element 𝑣 ∈ 𝑉 is either a substation or a two-step spectral clustering controlled islanding algorithm transformer, the edge 𝑒 = ( 𝑣 , 𝑣 )∈ 𝐸 is a physical 𝑖 ,𝑗 𝑖 𝑗 is introduced, while using generator coherency as the sole cable between two nodes [18, 19], and 𝑤 is the associated constraint with minimum active power flow disruption edge weight. A multi-layer graph G consists of 𝑀 distinct objective to find a suitable ICI solution. In [10, 11], the graph layers 𝐺 , 𝑖 = 1,2, … , 𝑀 , where each distinct layer authors presented an islanding scheme with minimal active 𝐺 = {𝑉 , 𝐸 , 𝑤 } is a undirected and weighted graph 𝑖 𝑖 𝑖 power flow disruption using a constrained spectral composed over a common vertex set 𝑉 and particular edge embedded clustering technique, while satisfying the set 𝐸 with associated weights 𝑤 [20]. The sets 𝑖 𝑖 generator coherency constraints. However, these comprising the graph assume interest from an operational techniques disregard the effects of the bus voltage and physical point of view. The individual layers magnitude, and reactive power, which has a substantial characterize specific relationships among entities, such as impact on the dynamic coupling. In [12], a methodology the frequency similarity associated with each pair of the based on dynamic frequency deviations of both generator island’s buses, and the active and reactive power flow and non-generator buses, with respect to the system disruption. nominal frequency is presented. Overall the center of inertia A generic representation of the three-layer graph in concept has shown its advantages in various applications. power system is depicted in Fig. 1, where the first, second While there is an in-depth treatment of individual topics and third layers are associated to the frequency similarity, such as generator coherency, optimization, and the active active power flow, and reactive power flow, respectively. and reactive power graph-based models in the literature, These layers have the same nodes which represent the buses there is a dearth of information regarding these multiple in power networks, while the edges are associated to the topics in a single model [13-17]. frequency similarity between buses in layer one, the active In this paper, a multi-layer graph spectral clustering power flow between buses in layer 2, and the reactive controlled islanding (M-SCCI) algorithm for ICI solution power flow between buses in layer three. While the first of power systems is presented. In the first stage of the layer is a full weighted graph (all the nodes are connected algorithm, the frequency similarity of buses, the active with each other), the other layers have the same edges as power flow between buses, and the reactive power flow physical lines in power networks. between buses construct three different layers of the multi- layer graph. The frequency similarity among each pair of buses is evaluated using correlation among the bus 2.2 Dynamic generator coherency frequency components. To determine the number of islands, the modularity clustering is applied to the layer containing Following a sudden disturbance on the power grid, the frequency similarities among buses, which results “k” dynamic response of individual generators can be numbers of coherent buses or coherent groups of generators. determined by phase angles dissimilarity at the buses near The number of “k” cluster outcomes of this grouping serves to the generator. The frequencies that represent the dynamic as the input in the second stage of the M-SCCI algorithm response of every generator after grid disturbances can be that identifies island boundaries with minimal active and defined as: reactive power flow disruption. This technique is based on a multi-layer graph, whose common vertex set represents 𝑠 = ∫ ( ∆𝜃 ( 𝑡 )− ∆𝜃 ( 𝑡 ) )d𝑡 (1) 𝑖 ,𝑗 𝑖 𝑗 the buses, and the edges on individual layers represent where 𝜃 and 𝜃 denote the phase angles at bus i and bus 𝑖 𝑗 power system attributes that reflect the similarities among j, respectively; T is the observation time; and 𝑠 is the 𝑖 ,𝑗 the buses in term of the various modalities. These dissimilarity index between bus i and bus j. modalities include: ① frequency correlation coefficient The amount of energy observed or delivered by between buses; ② real power flow disruption; ③ generators in the power system can be reflected by their reactive power flow disruption. speed deviations [20]. Therefore, analyzing these frequencies, which represent the dynamic response of generators following a disturbance, can be helpful for 2 Graph theory approach to controlled islanding coherency determination. These frequency components can problem be extracted using the discrete Fourier transform (DFT) as follows: 2 . 1 Multi-layer graph models of power systems An electrical network is undirected graph 𝐺 ( 𝑉 , 𝐸 , 𝑤 ) power systems, this factor represents the association between two different electrical buses, as shown in (5). A larger C indicates a stronger connection or higher C,ij coherency between bus 𝑖 and bus 𝑗 . ∑ [(𝐹 ( 𝑓 ) −𝐹 )(𝐹 ( 𝑓 ) −𝐹 )] 𝑓 =1 𝑖 𝑖 ,𝑎𝑣𝑔 𝑗 𝑗 ,𝑎𝑣𝑔 𝐶 = (5) 𝐶 , 2 2 𝑛 𝑛 ∑ (𝐹 ( 𝑓 ) −𝐹 ) ×∑ (𝐹 ( 𝑓 ) −𝐹 ) 𝑖 𝑗 𝑓 =1 𝑖 ,𝑎𝑣𝑔 𝑓 =1 𝑗 ,𝑎𝑣𝑔 where 𝐶 is the correlation coefficient among buses 𝑖 𝐶 , and 𝑗 ; 𝑛 is the number of frequency components comprised in range; and 𝐹 is the average of the 𝑖 ,𝑣𝑔𝑎 frequency components of bus 𝑖 in the domain of inter-area oscillation modes. We define the correlation coefficient similarity matrix M , as a matrix that its components are CCSM equal to 𝐶 𝑖 , 𝑗 ∈ 𝑁 , namely, 𝐶 , 𝐵 𝐶 ⋯ 𝐶 𝐶 ,11 𝐶 ,1𝑁 ⋮ ⋱ ⋮ 𝑴 = [ ] (6) 𝐶 ⋯ 𝐶 𝐶 ,𝑁 1 𝐶 ,𝑁 𝑁 𝐵 𝐵 𝐵 Fig. 1 Three-layer graph of power networks 2 . 4 Reactive power similarity matrix Generally, in a power system, the voltage and the frequency are controlled by reactive power and active 2π𝑓𝑘 𝑁 −1 −j power, respectively. Therefore, considering reactive power ( ) ( ) 𝑁 𝐹 𝑓 = ∫ 𝜔 𝑘 e d𝑘 , 𝑓 = 0,1, … , 𝑁 − 1 (2) 𝑖 𝑖 and active power simultaneously in the islanding problem 𝜃 ( 𝑘 ) −𝜃 ( 𝑘 −1) 𝑖 𝑖 𝜔 ( 𝑘 )= (3) ∆𝑡 would result in more stable islands in terms of frequency and voltage. In order to accomplish the aforementioned where 𝜔 ( 𝑘 ) is the angular velocity of generator 𝑖 at time goal, the minimal power flow disruption, as shown in (7), instant 𝑘 ; 𝐹 ( 𝑓 ) is the Fourier transform of the angular can be utilized for controlled islanding as the objective speed; N is the number of samples in the waveform; and ∆𝑡 function. is the time interval between two consecutive samples, held constant throughout simulations. T min ( ∑ |𝑄 |) ( ) ( ) ( ) The vector-space 𝑭 = [𝐹 1 , 𝐹 2 , … , 𝐹 𝑁 ] and (7) 𝑖 𝑖 𝑖 𝑖 𝑉 ,𝑉 ⊂𝑉 1 2 𝑖 ∈𝑉 ,𝑗 ∈𝑉 1 2 𝑁 × 𝑁 dimension matrix F are formed as: 𝑭 = [𝑭 , … , 𝑭 , … , 𝑭 ] (4) where 𝑄 is the reactive power flow between bus 𝑖 and 1 𝑖 𝑁 where 𝑁 is the total number of buses in the power grid. 𝐵 bus 𝑗 . The phase angle and the amplitude of each frequency The controlled islanding problem with the above component in the angular velocity signal can be extracted objective function can be transformed into a graph-cut by using the DFT. Therefore, the correlation of the phase problem by defining a squared 𝑁 × 𝑁 adjacency matrix 𝐵 𝐵 angle oscillation of generator/non-generator buses can with elements |𝑄 |. Accordingly, a reactive power graph reveal the coherency of oscillations related to generators, similarity matrix M is defined as: which will be discussed in the next section. 𝑀 = 𝑄 , |𝑄 |+|𝑄 | 𝑖𝑗 = |𝑉 ||𝑉 ||𝐺 𝑐 𝑜 𝑠 ( 𝜙 − 𝜙 ) | 𝑖 ≠ 𝑗 𝑖 𝑗 𝑖 𝑗 { 2 (8) 2.3 Correlation coefficient similarity matrix 0 𝑖 = 𝑗 Pearson product-moment correlation coefficient C is a where V and V are the voltage amplitudes of nodes i and j, i j popular metric used to evaluate the strength of the respectively; 𝐺 is the real part of the admittance matrix; association between two variables [5, 16]. The C ranges and 𝜙 , 𝜙 denote the phase angles between the voltage 𝑖 𝑗 between −1.0 and +1.0 and quantifies the direction and and the current at the respective nodes. strength of the linear association between two multidimensional random variables. In connection with 𝑖𝑗 𝑖𝑗 𝑗𝑖 𝑖𝑗 𝑖𝑗 𝑖𝑗 𝑖𝑗 𝐶𝐶𝑆𝑀 𝑖𝑗 𝑖𝑗 𝑖𝑗 2 . 5 Active power similarity matrix connections between them and to look for the naturally occurring groups in a network regardless of the number size. Similar to the reactive power, the minimal active power Greedy optimization of modularity tends to form very fast flow disruption, as shown in (9), can be defined and utilized clustering. for controlled islanding. The modularity is defined as the number of edges falling within groups minus the expected number in an equivalent network with edges placed at random. The modularity, min ( ∑ |𝑃 |) (9) denoted by Q, is given by: 𝑉 ,𝑉 ⊂𝑉 1 2 𝑖 ∈𝑉 ,𝑗 ∈𝑉 1 2 1 𝑑 𝑑 𝑖 𝑗 𝑄 = ∑ [𝑤 − ] 𝛿 ( 𝐶 , 𝐶 ) 𝑖 𝑗 where 𝑃 is the active power flow between bus 𝑖 and bus (11) 2𝑚 2𝑚 𝑗 . The controlled islanding problem with the above where 𝑤 is the weight of the edge between 𝑖 and 𝑗 ; 𝑑 objective function can be similarly transformed to a graph- and 𝑑 are the degrees of the vertices 𝑖 and 𝑗 , cut problem by defining a squared 𝑁 × 𝑁 adjacency 𝐵 𝐵 respectively; 𝑚 is the total number of the edges; and 𝛿 - matrix with elements |𝑃 |. Accordingly, an active power function is 1 if nodes 𝑖 and 𝑗 are in the same community graph similarity matrix M is defined as: ( 𝐶 = 𝐶 ), otherwise, it is 0. The value of 𝑄 lies in the 𝑖 𝑗 range [-1,1]. The cluster structure can be searched precisely 𝑀 = 𝑃 , by checking the network divisions that have large |𝑃 |+|𝑃 | 𝑖𝑗 = |𝑉 ||𝑉 ||𝐵 ( 𝜙 − 𝜙 ) | 𝑖 ≠ 𝑗 𝑖 𝑗 𝑖𝑗 𝑖 𝑗 modularity values. { 2 ( 10) 0 𝑖 = 𝑗 The first step in evaluating coherency of buses of a power network at any point in time is to calculate the correlation where 𝐵 is the imaginary part of the network admittance 𝑖𝑗 coefficient among all the buses and form the correlation matrix. coefficient similarity matrix. Then, 𝑘 groups of coherent Utilizing the minimal power flow disruption as the buses can be achieved by applying modularity clustering on objective function minimizes the amount of load that must the correlation coefficient similarity matrix. be shed following system splitting. The three proposed similarity matrics 𝑴 , 𝑴 and 𝑴 , are calculated 𝑄 𝑃 3.2 Stage II: controlled islanding while preserving based on real-time power system data. The use of the coherent bus groups aforementioned simlarity matrices in one model, one can be In graph theory, spectral clustering treats the data anticipated that an appropriate combination of information clustering as a graph partitioning problem, which is included in the multiple graph layers would lead to an equivalent to minimizing weights of graph cuts. Further, the improved clustering, i.e., this will lead to more precise normalized cuts algorithm can be used to find the solution predictions on the location and extension of the island of to the normalized cuts problem. It substantially corresponds stability. to working with the eigenvectors and eigenvalues of the normalized graph Laplacian. The normalized graph Laplacian matrix L is of broad interests in the studies of 3 Controlled islanding via multi-layer spectral spectral graph theory and is deﬁned as: clustering while addressing generator’s coherency 1 1 2 2 𝑳 = 𝑫 ( 𝑫 − 𝑾 ) 𝑫 (12) 3.1 Stage I: coherency detection based on modularity where 𝑫 is the degree matrix, i.e., a diagonal matrix with clustering the vertex degrees along the diagonal that are defined as Based on the concept of tight coherency, the phase angles 𝑀 𝐷 = 𝐴 , 𝐴 is the component of the adjacency 𝑗 =1 of all buses in an area should have relatively the same matrix 𝑨 of 𝐺 ; and 𝑾 is the adjacency matrix. deviation. This can be assessed by calculating the We consider now the problem of clustering 𝑁 vertices, correlation between each pair of buses in the area using (5). 𝑉 = { 𝑣 , 𝑖 = 1,2, … , 𝑁 } of 𝐺 into 𝑘 distinct subsets 𝑖 𝐵 To identify coherency of buses, it is necessary to find such that the bus nodes in the same subset are similar, i.e., strongly connected groups of buses since groups that are they are connected by edges of large weights. Reference [21] strongly coupled tend to maintain synchronism. Online proved that all normalized Laplacian eigenvalues of a graph coherency detection based on modularity clustering lie in the interval [0, 2], and 0 is always a normalized algorithm will be used to achieve this purpose. It neither Laplacian eigenvalue, a property favorable in comparing requires a predefined number of groups nor a defining different graph layers. We note that the spectral clustering threshold value. The objective of this method is to separate algorithms can efficiently solve this problem. Precisely, we the network into groups of vertices that have weak concentrate on the algorithm suggested in [19], which 𝑖𝑗 𝑖𝑗 𝑖𝑗 𝐶𝐶𝑆𝑀 𝑠𝑖𝑛 𝑗𝑖 𝑖𝑗 𝑖𝑗 𝑖𝑗 𝑖𝑗 𝑖𝑗 𝑖𝑗 𝑖𝑗 solves the following trace minimization problem: {𝑊 , 𝑖 = 1,2, … , 𝑀 } of each individual graph layers {𝐺 , 𝑖 = 1,2, … , 𝑀 }, k, and 𝛼 . T T ( ) min 𝑡𝑟 𝑼 𝑳 𝑼 , 𝑠 . 𝑡 . 𝑼 𝑼 = 𝑰 (13) 𝑁 ×𝑘 𝑼 ∈𝐑 Step 4: calculate the normalized Laplacian matrix 𝑳 and the subspace illustration 𝑼 for each 𝐺 . where 𝑘 is the target number of clusters, and 𝑁 is the 𝑖 𝑖 Step 5: compute the graph Laplacian matrix 𝑳 with total number of vertices in the graph. (16). The clustering of the vertices in 𝐺 is then implemented 𝑁 ×𝑘 using the 𝑘 -means clustering algorithm to the normalized Step 6: compute 𝑼 ∈ 𝐑 . row vectors of the matrix 𝑼 . Step 7: normalize each row of 𝑼 to get 𝑼 . 𝑚𝑛𝑜𝑟 Given a multi-layer graph 𝐺 with M individual layers Step 8: let 𝒚 ∈ 𝐑 ( 𝑗 = 1,2, … , 𝑛 ) be the transpose { } 𝐺 , 𝑖 = 1,2, … 𝑀 , we ﬁrst compute the graph Laplacian 𝑖 of the j-th row of 𝑼 . 𝑛𝑜 matrix 𝑳 for each 𝐺 and then represent each 𝐺 by the 𝑖 𝑖 𝑖 Step 9: cluster 𝒚 into 𝐶 , 𝐶 , … , 𝐶 using the 𝑘 - 𝑗 1 2 𝑘 𝑁 ×𝑘 spectral embedding matrix 𝑼 ∈ 𝐑 from the ﬁrst k means algorithm. eigenvectors of 𝑳 . Step 10: output cluster assignments 𝐶 , 𝐶 , … , 𝐶 . 1 2 𝑘 The goal is to merge these multiple subspaces in a meaningful and efficient way. To merge these multiple This algorithm uses the correlation coefficient between subspaces, the Riemannian squared projection distance the frequency components among 𝐶 , 𝑀 , and 𝑀 𝐶 , 𝑃 , 𝑄 , between the target representative subspace U and the M data to produce an islanding solution with minimal power { } individual subspaces 𝑼 , 𝑖 = 1,2, … , 𝑀 is computed as flow disruption. In the first stage, the buses are grouped the sum of the squared projection distances between U and using modularity clustering, based on the 𝐶 . The 𝐶 , each individual subspace given by 𝑼 : number of k clusters outcomes of this grouping serve as the 2 𝑀 𝑇 input to the second stage, in which nodes are grouped based 𝑑 = ∑ [𝑘 − 𝑡𝑟 ( 𝑼 𝑼 𝑼 𝑼 ) ] 𝑝𝑟𝑜𝑗 𝑖 =1 𝑖 𝑀 on multi-layer constrained spectral clustering. The M-SCCI 𝑇 𝑇 [( ) ] = 𝑘𝑀 − ∑ 𝑼 𝑼 𝑼 𝑼 ( 14) 𝑖 𝑖 algorithm proposed here can identify, in real time, an 𝑖 =1 islanding solution that has minimal power flow disruption By solving the following optimization problem that and satisfies the bus coherency constraints. integrates both (13) and (14), multiple subspaces can be merged. This method is based on the following Rayleigh- Ritz theorem, which transforms the generalized eigenvalues 4 Simulation studies problem into a constrained minimization problem, described as: The model effectiveness is evaluated through the 𝑀 𝑀 simulation study conducted on the modified IEEE 39-bus 𝑇 𝑇 system. The methodology has been implemented in 𝑚𝑛𝑖 𝑡𝑟 [𝑼 (∑ 𝑳 − 𝛼 ∑ 𝑼 𝑼 ) 𝑼 ] 𝑠 . 𝑡 . 𝑼 𝑼 = 𝑰 𝑖 𝑖 𝑁 ×𝑘 𝑼 ∈𝐑 MATLAB and all time-domain simulations are achieved in 𝑖 =1 𝑖 =1 DIgSILENT PowerFactory. To stress the system and raise ( 15) the likelihood of instability following a disturbance, we where 𝛼 is the weighting parameter that balances the increased the base load level by 25% at 0.01 s. Then, two trade-off linking the two terms in the objective function. We short circuit events occurred in lines 13-14 and 16-17 at 2 may note that this is identical trace minimization problem s. The short circuit events are cleared after 0.20 s by as introduced in (13), but with a “modified” Laplacian opening the line switches from the substations, while the given as: simulation lasts for 5 s. 𝑀 𝑀 Figures 2 and 3 demonstrate the rotor angle of generator and the system frequency, respectively, that indicate the 𝑳 = ∑ 𝑳 − 𝛼 ∑ 𝑼 𝑼 ( 16) 𝑚 𝑖 𝑖 system instability following the short circuit events. The 𝑖 =1 𝑖 =1 proposed solution approach is applied to the system to The proposed M-SCCI algorithm is described as follows. determine the islanding boundaries. The quality of each 1) First stage island is then evaluated by calculating the dynamic Step 1: formulate the multi-layer graph 𝐺 using only behavior and the power mismatch in the islands. It can be bus nodes, with edge weights equal to the 𝐶 , 𝑀 and observed from Fig. 3, that if no control action is undertaken, 𝐶 , 𝑃 , the system loses synchronism at about 2.25 s. Indeed, real- 𝑀 . 𝑄 , time simulation in DIgSILENT indicates out of step at 2.25 Step 2: obtain the 𝑘 cluster groups of coherent buses s for generators. As noticed, the system is divided into two from Step 1. groups, which are not balanced. 2) Second stage The frequency of the generators and the loss of Step 3: input 𝑁 × 𝑁 weighted adjacency matrices 𝐵 𝐵 𝑖𝑗 𝑖𝑗 𝑖𝑗 𝑖𝑗 𝑖𝑗 𝑖𝑗 𝑖𝑗 𝑟𝑚 synchronism are a clear indication that the system should groups of generators form the set of connectivity and non- be split. connectivity constraints, in which all pairs of generators in one group must be linked together (connectivity constraints), and generators in different groups must not be linked with each other (non-connectivity constraints), where the associated schematic is shown in Fig.4. Fig. 2 Rotor angle after two short circuit events without islanding Fig. 4 Coherent groups of generators and constraints of connectivity and non-connectivity After determining the coherent groups of generators, the active power graph similarity matrix is clustered into two groups, using the constraint clustering approach [22], where Fig. 3 System frequency after two short circuit events without islanding the clusters outcome and the rotor angles of generators following the clustering are depicted in Fig. 5, and Fig. 6, Before proceeding to discuss these case studies for our respectively. proposed methodology, we examine the islanding As can be seen, while the generators in island 2 are stable, methodology proposed in [9] to split the network. In this generators in island 1 become out of steps. On the other method, the authors proposed a two-step constrained hand, the graph-based islanding solutions automatically spectral clustering-controlled islanding to find the islanding satisfy the connectivity constraints, since the solution is solution, which provided the minimum power ﬂow sought through the minimum cuts in the graph. Our disruption while satisfying the constraint of coherent approach is a graph-based approach, wherein in each island generator groups. the nodes preserve their pre-islanding conditions, and the As it will be shown, in the following example, a multiple network is separated by cutting the edges in the graph. variant of valid cut-sets separating coherent generator G8 < 37 > groups from each other is possible, but only certain variants G10 < 30 > < 25 > will allow secure islanding. < 26 > < 28 > < 29 > According to the proposed model in [9], it is essential to < 27 > < 2 > < 38 > G9 find the minimum cut in a graph that its edges are the active < 24 > < 18 > < 17 > < 1 > power distortion and constraint the clusters with the G6 < 3 > < 16 > < 35 > coherent groups of generators. Accordingly, we need to first < 15 > G1 find the coherent groups of generators and then establish < 21 > < 22 > < 39 > < 4 > < 14 > the connectivity constraints among the generators within a < 5 > group and non-connectivity constraints between the < 12 > < 6 > < 19 > generators in different groups. Finally, the spectral < 23 > < 7 > constraint clustering is applied to the problem and < 13 > < 11 > < 36 > G7 determine the islands in the power system. < 10 > < 8 > < 20 > Following clearing the fault and applying the modularity < 31 > G2 < 32 > < 34 > < 33 > < 9 > clustering to the K matrix proposed in [16] at 2.21 s, two G5 G4 s G3 coherent groups of generators, {G1, G2, G3, G8, G9, G10} Fig. 5 Islanding boundaries after applying the clustering method and {G4, G5, G6, G7}, are produced. The two coherent G8 < 37 > G10 < 30 > < 25 > < 26 > < 28 > < 29 > < 27 > < 2 > < 38 > G9 < 24 > < 18 > < 17 > < 1 > G6 < 16 > < 3 > < 35 > < 15 > G1 Fig. 6 Rotor angles of generators after applying the clustering method < 22 > < 21 > < 39 > < 4 > < 14 > < 5 > The following sub-sections compare the result using < 12 > < 6 > three different criteria, i.e. frequency similarity, reactive < 19 > < 23 > < 7 > power, and active power, for the islanding decision making < 13 > < 11 > < 36 > G7 procedure. Four cases under the same operating conditions < 10 > < 8 > < 20 > < 31 > are employed to demonstrate the accuracy of the proposed G2 < 32 > < 34 > < 33 > < 9 > G3 G5 G4 M-SCCI algorithm. Case 1: single-layer intentional islanding based on Fig. 7 Islanding boundaries considering frequency similarity frequency similarity of the island’s buses. Case 2: single-layer intentional islanding based on Figure 9 shows the generator rotor angle oscillation reactive power. during the simulation study of 15 s. Obviously, the rotor Case 3: single-layer intentional islanding based on active angle oscillations are unstable, and all the machines lose power. synchronism while groups of generators become weaker Case 4: multi-layer intentional islanding based on all following the events. three criteria. In all cases, the islanding scheme is applied at t=2.21 sec, just after clearing the fault to avoid generator instability that happens at t=2.25 sec if no action is taken. 4.1 Case 1 In this case study, the frequency similarity is employed as the main criterion for islanding decision making. The approach provides a suitable islanding solution using online Fig. 8 Rotor angles after islanding based on frequency similarity coherency and pre-fault power flow conditions. In the first stage, the proposed buses coherency modularity clustering 4.2 Case 2 algorithm based on frequency similarity of the island’s buses identified two sets of coherent generators {G1, G2, In this case study, the minimum reactive power flow G3, G10} and {G4, G5, G6, G7, G8, G9}. disruption is employed as the main criterion for islanding Considering these two sets found in first stage, the decision making procedure. The system initial condition is number of two clusters outcomes serves as the input in the the same as that of Case 1. The same faults as that of Case second stage to solve the single-layer constrained spectral 1 are imposed. In the first stage of the proposed bus clustering. The islanding solution suggests that it should be coherency, modularity clustering algorithm based on split into two islands as shown in Fig. 7. The resulted reactive power similarity of the island’s buses identified groups using 𝐶 are two groups as depicted in Fig. 8 𝐶 , three sets of coherent generators {G1, G2, G3, G10}, {G4, with two background colors. G5, G6, G7}, and {G8, G9}. The allocation of buses in Case 1 to coherent generator Considering these three sets found in the first stage, the groups is as follows: ① island 1, buses B1-B14, B30-B32, number of three clusters outcomes serves as the input in the and B39; ② island 2, buses B15-B29 and B33-B38. The second stage to solve the single-layer constrained spectral minimal power flow disruption across boundaries of islands clustering based on the minimum reactive power flow are 857 MW active power and 1349 Mvar reactive power. 𝑖𝑗 Fig. 9 Correlation coefficient similarity matrix at 2.21 s power. disruption. The islanding solution suggests that it should be Figure 12 shows the generator rotor angle oscillation three islands as shown in Fig. 10. The resulted groups using during the simulation study of 15 s. Obviously; the rotor 𝑀 are three groups as depicted in Fig. 11 with three angle oscillations are damped, and all the machines lose 𝑄 , synchronism while groups of generators became weaker background colors. following the events. G8 < 37 > G10 < 30 > < 25 > < 26 > < 28 > < 29 > < 27 > < 2 > < 38 > G9 < 18 > < 17 > < 24 > < 1 > G6 < 16 > < 3 > < 35 > < 15 > G1 < 22 > < 21 > < 39 > < 4 > < 14 > < 5 > < 12 > < 6 > < 19 > < 23 > < 7 > < 13 > < 11 > Fig. 11 Rotor angles after islanding based on reactive power < 36 > G7 < 10 > < 8 > < 20 > < 31 > 4.3 Case 3 < 32 > G2 < 34 > < 33 > < 9 > G3 G5 G4 In this case study, the minimum active power flow Fig. 10 Islanding boundaries considering reactive power disruption is employed as the main criterion for islanding The allocation of buses in Case 2 to coherent generator decision making procedure. The system initial condition groups is as follows: ① island 1, buses B1-B15, B18, and the fault are the same as that of Case 1. The ﬁrst stage of the proposed bus coherency B30-B32, and B39; ② island 2, buses B16, B19-B24, and modularity clustering algorithm based on active power B33-B36; ③ island 3, buses B17, B25-B29, B37, and B38. similarity of the island’s buses returned three coherent The minimal power flow disruption across boundaries of islands are 2291 MW active power and 1349 Mvar reactive 𝑖𝑗 Fig. 12 Reactive power similarity matrix at 2.21 s groups is as follows: ① island 1, buses B1-B15, B25, generator groups {G1, G2, G3, G8, G10}, {G4, G5, G6, B30-B32, and B39; ② island 2, buses B16-B24 and B33- G7}, and {G9}. B36; ③ island 3, buses B26-B29 and B38. The minimal G8 power flow disruption across boundaries of islands are < 37 > G10 1108 MW active power and 1349 MVar reactive power are < 30 > < 25 > < 26 > < 28 > disrupted. < 29 > < 27 > < 2 > < 38 > Figure 15 shows the generator rotor angle oscillation G9 < 24 > < 18 > < 17 > during the simulation study of 15 s. Obviously, the rotor < 1 > G6 < 16 > < 3 > angle oscillations are damped, and all the machines lose < 35 > < 15 > synchronism. As can be seen, the power system is not stable. G1 < 21 > < 22 > < 39 > < 4 > < 14 > < 5 > < 12 > < 6 > < 19 > < 23 > < 7 > < 13 > < 11 > < 36 > G7 < 10 > < 8 > < 20 > < 31 > G2 < 32 > < 34 > < 33 > < 9 > G3 G5 G4 Fig. 13 Islanding boundaries considering active power Considering these three sets found in the first stage, the number of 3 clusters outcomes serves as the input in the Fig. 14 Rotor angles after islanding based on active power second stage to solve the single-layer constrained spectral clustering based on the minimum active power flow disruption. The resulted groups, using the 𝑀 are three 𝑃 , groups as depicted in Fig. 13 with three background colors. The islanding solution suggests that there should be three islands as shown in Fig. 14. The allocation of buses in Case 3 to coherent generator 𝑖𝑗 Fig. 15 Active power similarity matrix at 2.21 s G8 < 37 > G10 4.4 Case 4 < 30 > < 25 > < 26 > < 28 > < 29 > < 27 > < 2 > < 38 > The system initial condition and the fault are the same G9 < 24 > < 18 > < 17 > as all the previous cases. In this case study, the frequency, < 1 > G6 < 16 > active power, and reactive power similarity matrices are < 3 > < 35 > < 15 > employed as the main criteria for islanding decision making G1 < 21 > < 22 > procedure. We implemented intentional islanding at 2.21 s < 39 > < 4 > < 14 > following two cascading outages. First, correlation < 5 > coefficient is calculated using (5) for all pairs of buses < 12 > < 6 > < 19 > which result in the 𝐶 shown in Fig. 8. Applying the 𝐶 , < 23 > < 7 > modularity clustering on the 𝐶 returned two coherent < 13 > 𝐶 , < 11 > < 36 > G7 generator groups {G1, G2, G3, G8, G9, G10} and {G4, G5, < 10 > < 8 > < 20 > G6, G7}. < 31 > < 32 > G2 < 34 > < 33 > < 9 > G3 G5 G4 Considering these two sets found in the first stage, the Fig. 16 Islanding boundaries based on multi-layer clustering number of two clusters outcomes serves as the input in the second stage to solve the M-SCCI. Then, in the second The final splitting strategy, possessing the lowest power stage, the M-SCCI algorithm was excused using the three- exchange is represented in Fig. 16. The minimal power flow layer graph with weighted adjacency matrices, i.e. 𝑴 , disruption across boundaries of islands are 622 MW active 𝑴 and 𝑴 , as the main criteria for islanding decision power and 1349 Mvar reactive power. Figure 17 shows the 𝑄 𝑃 generator rotor angle oscillation during the simulation study making procedure, taking into consideration the two cluster of 15 s. Obviously, the rotor angle oscillations are damped, coherency groups found in the first stage of the algorithm. and all the machines remain in synchronism while groups The allocation of buses in Case 4 to the coherent generator of generators became stronger following the events. Table groups is as follows: ① island 1, buses B1-B14, B17, B18, 1 presents the power flow mismatch between the islands for B25-B32, and B37-B39; ② island 2, buses B15, B16, each case study. B19-B24, and B33-B36. The comparison in Table 1 shows that the proposed M- SCCI algorithm using all criteria returns the cut-set that separated the coherent generator groups with minimum cut, 𝐶𝐶𝑆𝑀 𝑖𝑗 𝑖𝑗 which is 622 MW. Table 1 Summary of power flow mismatch between islands for each case study Active Active Reactive Reactive power power power power ∆P ∆Q ∑|∆P| ∑|∆Q| Clustering criteria Island (generator) (load) (generator) (load) ( ) ( ) MW MVar ( MW) ( Mvar) P P Q Q G L G L ( MW) ( MW) ( MVar) ( MVar) 1 3672 4765 1663 1458 -1092 205 Clustering based on reactive 2 3525 2681 1397 364 844 1032 2291 1349 power 3 2055 1700 402 291 355 111 1 4482 4864 1802 1484 -381 318 Clustering based on active 2 3525 2918 1397 409 607 987 1107 1349 power 3 1245 1364 264 221 -119 44 1 5580 5098 1799 930 482 869 Clustering based on frequency 857 1349 similarity 2 3672 4048 1663 1183 -375 480 1 2055 1700 402 291 355 111 Clustering based on all three 2 3525 3161 1397 594 364 803 1331 1349 criteria, k=3 3 3672 4285 1663 1228 -612 435 1 5727 5985 2066 1520 -258 546 Clustering based on all three 622 1349 criteria, k=2 2 3525 3161 1397 594 364 803 6 References [1] Trodden PA, Bukhsh WA, Grothey A et al (2014) Optimization- based islanding of power networks using piecewise linear AC power flow. IEEE Trans Power Syst 29(3):1212-1220 [2] Sun K, Zheng D, Lu Q (2003) Splitting strategies for islanding operation of large-scale power systems using OBDD-based methods. IEEE Trans Power Syst 18(2):912-923 [3] Ding T, Sun K, Huang C et al (2018) Mixed-integer linear programming-based splitting strategies for power system islanding operation considering network connectivity. IEEE Fig. 17 Rotor angles after islanding based on all criteria Systems Journal 12(1):350-359 [4] Trodden P, Bukhsh W, Grothey A et al (2013) MILP formulation for controlled islanding of power networks. International Journal of Electrical Power and Energy Systems 45(1):501-508 [5] Kamali S, Amraee T (2017) Blackout prediction in 5 Conclusion interconnected electric energy systems considering generation re-dispatch and energy curtailment. Appl Energy 187:50-61 [6] Kamali S, Amraee T, Capitanescu F (2018) Controlled network This paper proposed a computationally efficient real- splitting considering transient stability constraints. IET time ICI algorithm based on multi-layer graphs, subspace Generation Transmission and Distribution 12:5639-5648 analysis, and constrained spectral clustering while [7] Ding L, Ma Z, Wall P et al (2017) Graph spectra based controlled addressing the generator coherency problem. We islanding for low inertia power systems. IEEE Trans Power Del 32(1):302-309 demonstrated that using multi-layer spectral clustering to [8] Chong Z, Dai Z, Wang S et al (2014) The application of binary find the islanding boundaries, instead of using a single layer, particle swarm optimization in power restoration. In: i.e., the frequency similarity, the active power, and the Proceedings of 10th international conference on natural reactive power produced improved clustering performance. computation (ICNC), Xiamen, China, 19-21 August 2014, pp The insertion of the bus coherency constraints prevents new 349-353 [9] Ding L, Gonzalez-Longatt FM, Wall P et al (2013) Two-step island groupings that would contain non-coherent spectral clustering controlled islanding algorithm. IEEE Trans generators. The use of minimal power-ﬂow disruption Power Syst 28(1):75-84 improves the transient stability of the islands produced. The [10] Esmaeilian A, Kezunovic M (2016) Controlled islanding to simulation results show that the proposed M-SCCI prevent cascade outages using constrained spectral k-embedded clustering. In: Proceedings of power systems computation algorithm is computationally efficient and is suitable for conference (PSCC), Genoa, Italy, 20-24 June 2016, 6 pp use in real-time applications involving large power systems. [11] Quirós-Tortós J, Wall P, Ding L et al (2014) Determination of sectionalising strategies for parallel power system restoration: a spectral clustering-based methodology. Electr Power Syst Res 116:381-390 [12] Khalil AM, Iravani R (2016) A dynamic coherency identification method based on frequency deviation signals. IEEE Trans Power Syst 31(3):1779-1787 [13] Shao H (2016) Adaptive three-stage controlled islanding to prevent imminent wide-area blackouts. Dissertation, Durham University [14] Davarikia, H., Znidi, F., Aghamohammadi, M. R., & Iqbal, K. (2016, July). Identification of coherent groups of generators based on synchronization coefficient. In 2016 IEEE Power and Energy Society General Meeting (PESGM) (pp. 1-5). IEEE. [15] Znidi, F., Davarikia, H., & Iqbal, K. (2017, July). Modularity clustering based detection of coherent groups of generators with generator integrity indices. In 2017 IEEE Power & Energy Society General Meeting (pp. 1-5). IEEE. [16] Davarikia, H., Barati, M., Znidi, F., & Iqbal, K. (2018, August). Real-time integrity indices in power grid: a synchronization coefficient based clustering approach. In 2018 IEEE Power & Energy Society General Meeting (PESGM) (pp. 1-5). IEEE. [17] Davarikia, H., & Barati, M. (2018). A tri-level programming model for attack-resilient control of power grids. Journal of Modern Power Systems and Clean Energy, 6(5), 918-929. [18] Dong X, Frossard P, Vandergheynst P et al (2014) Clustering on multi-layer graphs via subspace analysis on Grassmann manifolds. IEEE Transactions on Signal Processing 62(4): 905- [19] Dong X, Frossard P, Vandergheynst P et al (2012) Clustering with multi-layer graphs: a spectral perspective. IEEE Transactions on Signal Processing 60(11): 5820-5831 [20] Jonsson M, Begovic M, Daalder J (2004) A new method suitable for real-time generator coherency determination. IEEE Trans Power Syst 19(3):1473-1482 [21] Chung FR, Graham FC (1997) Spectral graph theory. American Mathematical Society, Providence [22] Rangapuram SS, Hein M (2012) Constrained 1-spectral clustering. In: Proceedings of the 15th international conference on artificial intelligence and statistics (AISTATS), La Palma, Canary Islands, 21-23 April 2012, 12 pp

Electrical Engineering and Systems Science – arXiv (Cornell University)

**Published: ** May 29, 2019

Loading...

You can share this free article with as many people as you like with the url below! We hope you enjoy this feature!

Read and print from thousands of top scholarly journals.

System error. Please try again!

Already have an account? Log in

Bookmark this article. You can see your Bookmarks on your DeepDyve Library.

To save an article, **log in** first, or **sign up** for a DeepDyve account if you don’t already have one.

Copy and paste the desired citation format or use the link below to download a file formatted for EndNote

Access the full text.

Sign up today, get DeepDyve free for 14 days.

All DeepDyve websites use cookies to improve your online experience. They were placed on your computer when you launched this website. You can change your cookie settings through your browser.