Electrical Power and Energy Systems 35 (2012) 138147

Contents lists available at SciVerse ScienceDirect

Electrical Power and Energy Systems

Intentional islanding using a new algorithm based on ant search mechanism

a r t i c l e

i n f o

a b s t r a c t
Cascading failures and blackouts are the most signicant threats for power system security. If the process of cascading failure proceeds by further line tripping, the system will face uncontrolled islanding. Establishment of uncontrolled islands with deciency in MW or Mvar power balance are the main reasons for system blackout. In order to reduce the risk of blackout due to islanding, intentional or controlled islanding has been considered as a preventive strategy. In this paper, for identifying proper islanding scenarios in network, a new search algorithm based on the ant search mechanism is proposed. The security constraints considered for nding islanding scenarios are loadgeneration balance and line overloading which are implemented using linear programming and DC load ow. The proposed algorithm has been applied on IEEE 39-bus network with promising results showing the ability of the algorithm for nding proper islanding scenarios quickly. 2011 Elsevier Ltd. All rights reserved.

Article history: Received 30 January 2011 Received in revised form 13 September 2011 Accepted 10 October 2011 Available online 5 November 2011 Keywords: Blackout Intentional islanding Power balance Line overloading Load shedding

1. Introduction Security is a vital requirement for power systems which can be dened as its ability for preventing blackout following initiation of cascading failures. The process of cascading failures is the main mechanism for pushing power system toward blackout. Cascading failure is a complicated process consisting of a sequence of events such as line tripping, which take place due to network weaknesses and local function of protection relays. In recent years, numerous ways for reducing cascading failure risks have been identied, including: more coordinated emergency controls. In general terms, these suggestions are attempts to improve coordination of power system design and operation to decrease cascading failures caused by line overloading. In [1], a decentralized load shedding approach has been presented that mimics wide-area approaches to provide emergency protection against excess frequency decline which also provides protection against line overloading, and hence minimizes cascading failure risk. A key feature of the proposed load shedding scheme is the use of local frequency rate information to adapt the load shedding behavior to the size and location of the experienced disturbance. Modeling cascading failures is crucial for analyzing power system blackouts. A malfunction of a power system shows usually itself as a blackout. In [2], a new approach for blackout modeling based on ignoring the details of particular failures and focusing on the study of global behaviors and dynamics of time series with

Corresponding author. Tel.: +98 21 77312176; fax: +98 21 73932591.

approximate global models has been presented. Concepts such as criticality and self-organization have been applied to characterize blackout data, suggesting that the frequency of large blackouts is governed by non-trivial distribution functions such as power laws. In the process of cascading failures, outage of a heavily loaded line or tripping of a large generator may lead the system to collapse immediately. It is very important to maintain power ow solvability when unsolvable cases occur. Load shedding can prevent a system blackout in these situations. In [3], a framework based on two sub-problems using LP based OPF has been presented for determining a load shedding strategy for the restoration of power ow solvability and improvement of VSM. In the process of cascading failure, if line tripping proceeds further, power network splitting into uncontrolled islands will be inevitable. The uncontrolled islands always suffer from load generation imbalance which may cause angle, voltage or frequency instability leading to blackout. In order to prevent cascading outages, in [4] a technique based on graph theory has been presented for generation shift from one generator to another. This method is suitable for both credible and non-credible contingencies. The proposed method also provides important additional information i.e. the transmission lines (links) to be switched off to detect islands of a power system in addition to preventive control strategies. Forced establishment of uncontrolled islands is dominant characteristic of power system dynamic during the process of cascading failures which is recognized as the main cause for blackout. In order to avoid uncontrolled separation of power network and reducing the risk of blackout, intentional islanding has been considered as a preventive action in defense strategy of power systems.

M.R. Aghamohammadi, A. Shahmohammadi / Electrical Power and Energy Systems 35 (2012) 138147

139

System splitting known as controlled islanding is intentional separation of entire network into two or several islands by tripping properly selected lines. After system splitting, the whole power system will be under islanding operation and each island including its own load and generation should be able to preserve its balance and remain stable. In such situation, although the power system is operating in an abnormal degraded state, however, customers are continuing to be served [5]. In order to apply the strategy of controlled islanding, three tasks should be carried out sequentially. 1. Recognizing the proper operating instant at which applying intentional splitting is inevitable otherwise the system will be separated into uncontrolled islands. 2. Identifying the proper islands for intentional network separation such that each island will be able to preserve its power balance and stability. 3. Implementation of the planned islanding scenario in a proper way without any dynamic and transient consequence causing large oscillation and instability for islands. However, regarding tasks 1 and 3, less works have been reported [6]. When islanding scenario is identied and decided, the most important task is implantation of the scenario by proper tripping of lines between islands. In this respect, the order of line tripping is very important and dominating for islands stability. Most reported works are mainly focused on task 2 concerning identication of proper islanding scenarios for network splitting [711]. Detecting islands and determining asynchronous groups of generators have been investigated in [1214]. Enhancing the functions and operation schemes of relays for reducing their contribution in system blackout has been worked in [11,15]. However, still for recognition of controlled islanding strategies to prevent blackouts, limited studies have been reported [5,1619]. Algorithms based on the technique Ordered Binary Decision Diagram (OBDD) have been used to determine proper splitting strategies [5,16,17,20]. In [16], a three phase method has been used to determine proper islanding scenario in the network. In Phase-1, a much simpler reduced network of the original power network is constructed by graph theory; then in Phase-2, the verication algorithms based on OBDDs can efciently narrow down the strategy space and can give enough splitting strategies satisfying loadgeneration balance constraints. In Phase-3, by using power-ow calculations the possibility of line overloading is checked, and nal proper splitting strategies will be given. Also in [5] a two phase method has been used to nd proper islanding schemes. The method narrows down the strategy space using highly efcient OBDD-based algorithm in the rst phase, then nds proper splitting strategies using power-ow analysis in the reduced strategy space in the second phase. In [17], in addition to loadgeneration balance constraint and overloading constraint in transmission lines, some other constraints such as synchronism between generators in each island and stability of islands after splitting are considered for nding proper islands. In this reference, a three phase method has been used to satisfy these constraints in order to nd proper islands. In the rst phase, power network will be split into separated sub networks; in the second phase, the constraints of loadgeneration balance and groups of synchronized generators will be checked. Finally in the third phase, constraints of overloading in transmission lines and stability of separated islands will be checked. The main aims of network islanding can be listed as follows; 1. Isolating the vulnerable parts from other parts in order to avoid propagation of weak areas problems to other parts.

2. Splitting the whole system into small subsystems for easy handling in dynamic and emergency conditions. After splitting, each island including some loads and generators, must be able to maintain its stability and power balance by controlling its generators in a permissible limit and applying load curtailment. In this paper, a new algorithm based on ant search mechanism is proposed for identifying controlled islanding scenarios. Here, it is assumed that the rst task has been done and decision for islanding has been taken; so concerning the second task, and in order to identify proper islanding scenario, this approach is proposed. The third task is not considered in this paper and it is the subject of another study.

2. Intentional and controlled islanding In the process of power system dynamic caused by cascading failures, some parts of network may experience angle, frequency or voltage instability. In such situation, trying to maintain system integrity and operate the system entirely interconnected is very difcult and may cause propagation of local weaknesses to other parts of the system. In critical conditions which continuation of cascading failures threats integrity of whole network and may split it into uncontrolled islands, intentional separation of power system into controlled islands is recognized as an effective defense strategy. Intentional controlled islanding of power systems is based on the following advantages: 1. Separating weak and vulnerable areas from stable parts of the system. 2. Easy handling and control of small subsystems with respect to the whole system. After formation of islands, loadgeneration imbalance, line overloading, voltage, angle and frequency instabilities, are the phenomena which can threat the stability and integrity of each island. Therefore, islands must be formed in such way to be able to manage and maintain static and dynamic stability of their own region independently. For this purpose, each island must have adequate resources of active and reactive power and sufcient control facilities like load shedding and generation reserve. The number of islands that should be established in the network is better to be as less as possible, however, it depends on the system condition at the instant of islanding, including network structure, spread of vulnerability into network, slow coherent groups of generators and available control facilitates in the network. As mentioned above, the process of intentional islanding consists of three phases including decision making for islanding, identifying proper islanding scenario and implementing islanding scenario. Decision depends on the criticality of operational conditions. Whenever system operator recognizes system inability for preserving its integrity, a proper islanding strategy can be applied as a defensive solution. In this paper, based on the hypothesis of having decided to split the network into controlled islands, a probabilistic algorithm for identifying the proper islanding scenario is proposed. The proposed approach is able to identify a variety of islanding scenarios satisfying basic criteria. In addition, it enables operator to implement some other constrains like minimum number of boundary lines between islands which cannot be easily handled in the main process of island identication. Also, provision of variety of islanding scenarios enables operator to effectively implement his engineering understanding and experience in the process of adoption of proper islanding scenarios.

140

M.R. Aghamohammadi, A. Shahmohammadi / Electrical Power and Energy Systems 35 (2012) 138147

3. Criteria for intentional islanding After decision for network islanding, the most important step is identifying the proper stable islanding scenarios. The criteria for nding proper stable islanding scenarios can be categorized as follows; 1. Integrity criterion: all buses inside an island must be connected as an integrated subsystem. 2. Static criteria: following island formation, at the steady state condition, the following criteria should be satised in each island which guarantee static stability of islands;  Criterion 1 Ability to preserve loadgeneration balance including generation control and load shedding capacities in each island. Eqs. (1) and (2) represent island ability for preserving active power balance in the cases of load rich or generation rich respectively.
n n X X PGri PLsi P PLi PGi > 0 i1 i1

ibility for implementing any kind of system constrains into process of island identication. For example, although, due to DC load ow modeling of power system, voltage constraint has not been directly considered as a criterion, but the algorithm is able to handle reactive loadgeneration balance as further constraint like Eqs. (2) and (3). The proposed approach can be implemented in a control center as a tool for the emergency condition to nd and suggest several preliminary feasible islanding scenarios with possibility to include any further constraints and requirements. The process of the proposed algorithm in this paper can be explained through four steps as follows. 4.1. Structural integrity The rst and the most important step in the process of identifying islanding scenarios is to nd all possible splitting schemes in the network satisfying the integrity criterion. The proposed algorithm for nding splitting schemes in the network is based on a search mechanism denoted as ant search algorithm combined with a probabilistic movement mechanism. 4.2. Loadgeneration balance

n n X X PGri P PGi P Li > 0 i1 i1

2
In this step, the ability of each splitting scheme for preserving loadgeneration balance is veried using Eqs. (1) or (2) with respect to the available generation reserve and load shedding capacities. Each splitting scenario not satisfying this constraint even for only one of its separated areas will be terminated and eliminated. This criterion is implemented within the rst step, so the integrity and power balance criteria are veried simultaneously. 4.3. Line overloading constraint In this step, for splitting schemes fullled steps 1 and 2, the ability of each separated area for preventing line overloading is veried. For this purpose, the available generation reserve capacity for generation rescheduling and load shedding capacity are used as control parameters for evaluating the ability of each area to prevent line overloading. Verication of this step is carried out by means of linear programming using Eqs. (3)(7) which are based on DC load ow equations.

where PGri is the controllable generation reserve capacity at bus i which can be used for generation increase/decrease, PLsi the load shedding capacity at bus i, PGi the Active generation power at bus i, PLi the active load power at bus i, and n is the number of buses in the island.  Criterion 2 Line overloading constraint for all lines of the island.  Criterion 3 Voltage drop constraint for all buses of the island.

 Criterion 4 Voltage security margin constraint. 1. Dynamic criteria: At the transient period of islands formation due to tripping boundary lines, the following criteria should be satised in each island which guarantee dynamic stability of the islands  Generators should remain in synchronism with damped oscillation.  Voltage should remain stable dynamically.  Frequency should remain stable within acceptable limit. In this paper, for nding stable islanding scenarios, the integrity and the two rst static criteria are considered as basic criteria for satisfaction. For this purpose, a new algorithm based on ant search mechanism is proposed. It is noteworthy that for practical implementation of an islanding scenario all static and dynamic criteria should be fullled. Therefore, the islanding scenarios found by this algorithm could be used as preliminary cases for nding nal scenario. 4. Proposed approach It should be noted that practical implementation of an islanding scenario requires satisfaction of all criteria. For this purpose, as the search process proceeds further with respect to the fulllment of more criteria, the search space becomes narrower until it reaches to nal scenarios. Therefore, the proposed algorithm in this paper is able to nd several number of islanding scenarios with respect to the integrity and the two rst static criteria. In practical environment, these scenarios can be used as preliminary feasible scenarios for further investigation, until the point that all criteria will be fullled in the nal scenario including operator experience and knowledge. The advantageous of the proposed approach is its ex-

Min :L s:t: :

n X i1

P Lsi

n n X X PGi PGri PLi PLsi i1 i1 n X i1

4 5 6 7

FK

C kJ Pj 6 F max K

8 K 1; 2; . . . ; m

Pmin 6 PGri 6 Pmax Gri Gri PLsi 6 Pmax Lsi

where PGi is the active generation at bus #i at the moment of islanding, PLi the active power load at bus #i at the moment of islanding, PGri the active generation control at bus #i, PLsi the load shedding at bus #i, FK the active power ow of line #K, F max the maximum K power capacity of line #K, P max > 0 the maximum available generaGri min tion increase at bus #i, PGri < 0 the maximum available generation decrease at bus #i, Pmax the capacity of load shedding at bus #i, C the Lsi matrix relating line active power ow to bus active power injection, and m is the number of lines. In this step, the ability of each separated area belonging to an islanding scenario is veried individually for preventing line overloading. Even if only one area failed to prevent line overloading, the

M.R. Aghamohammadi, A. Shahmohammadi / Electrical Power and Energy Systems 35 (2012) 138147

141

corresponding islanding scenario will be terminated and eliminated from search space. In this algorithm, the objective function is merely based on the minimization of load shedding but generation control is set free within its control limits. Eqs. (4) and (5) represent power balance and line overloading constraints based on DC load ow equations respectively. 4.4. Final verication After nishing step 3, those splitting schemes which have fullled all criteria will be selected as acceptable preliminary islanding scenarios. For each islanding scenario, the boundary lines connecting islands, which must be tripped to establish isolated islands, are recognized. Power ow of boundary lines, between each two islands before splitting, is calculated. In this step, in addition to dynamic criteria, some additional criteria can be dened as follows to merely verify easy and secure implementation of islanding scenarios:  Number of boundary lines connecting separated islands.  Power ow of boundary lines before splitting. It is clear that the islanding scenario with less number of boundary lines and less power transfer are easier for implementation. However, it is understood that implementation of an islanding scheme is a very complicated process which requires different dynamic and protective considerations in the network. For example, the following points are part of problems which must be fully considered. 1. The order of islands to be separated from other parts of network. 2. The order of boundary lines for tripping. 3. The control and protection actions which should be taken. 5. Principles of search algorithm In this paper, principle of a probabilistic search mechanism denoted as ant search algorithm is developed for nding different islanding scenarios. In this algorithm, for each islanding scenario, the number of islands can be adopted prior to the search. However, for effective search the number of islands should be adopted based on engineering and practical considerations. The most important dynamic constraint is number of potential coherent groups of generators in the system. The most practical consideration is easy handling and establishment of islands. Regarding this fact, the less the number of islands is, the higher the search speed and easier the establishment of islanding scenario would be. In this approach, the search for nding each islanding scenario starts simultaneously and in parallel from a number of randomly selected initial points equal to the number of intended islands. The search algorithm can be carried out many times and in each time, the algorithm may either satisfy all criteria or not. In the case of satisfaction of all criteria, one islanding scenario will be found. Principles of the proposed search algorithm can be explained in three steps as follows. Fig. 1 shows ow chart of the proposed search algorithm. 5.1. Pre-processing network structure In order to speed up the search process for nding islanding scenarios, it is necessary to reduce the network size by removing unnecessary points (buses). For this purpose, a group number (GN) is assigned to each point. The default GN for all buses is 0 indicating no dependency between them. The same non-zero GN for a group of buses means that they are dependent as one point and must locate in the same island. For example, the point at the

end of a radial line takes the same GN as the rst point. The points intended to be located in the same island take the same individual GN number rather than 0. Also for easy handling generators coherency into the process of island search, in the case of recognizing coherent groups of generators, all generator buses belonging to one coherent group can be assigned with the same GN.

5.2. Adoption of initial points for search The search for nding each islanding scenario starts simultaneously and in parallel from a number of randomly selected initial nodes equal to the number of intended islands. The location of initial nodes are adopted based on network structure and operational constraints. For example, in the case of existing coherent groups of generators, the initial points should be selected within the area of each coherent group. In this paper, in order to increase the probability of search convergence and avoiding interference of search activities for islands, initial points are selected randomly in sequence. After random selection of each initial point, a corresponding forbidden area is dened around that point such that the next initial point is not allowed to locate within the forbidden areas of other initial points. The forbidden area corresponding to each initial point consists of buses connected directly or indirectly to that point. The approximate size of the forbidden area is determined based on the size of whole network and the number of intended islands from the following equation:

Ni

Nbus Nisland

where Nbus and Nisland are number of total buses and intended islands respectively. Ni is approximate number of buses in the forbidden area corresponding to each intended island.

5.3. Ant movement for nding islands After choosing the initial points, from each point an ant starts to search for nding an island according to the following rules. In order to speed up the search activity, all criteria mentioned in Section 4 are implemented within the movement of ants for island identication. 1. Initial status of all buses is set to 0 indicating their availability for occupation by ants. 2. Based on the operational condition, the number of intended islands (Nisland) is decided with corresponding ants. 3. For each ant an ID number (K = 1, 2, . . . , Nisland) corresponding to the number of its intended island is assigned. 4. When a bus is occupied by ant K, its status changes to K indicating its occupation and attribution to Kth island. 5. Each ant only moves toward unoccupied points with status 0 connecting to current position of the ant. 6. When status of a bus changed to an ID (K), the status of all buses with the same GN number change to that ID indicating their location in the same island K. 7. The next point for each ant movement is randomly selected with highest probability from all unoccupied points connecting to current position of the ant. 8. An ant stops searching when there is no point with status 0 connecting to it. 9. After stopping ant K, all points occupied by it with the same ID constitute corresponding island K. 10. When all ants stop searching, the remained points with status 0 will be connected to the nearest islands subject to minimum loadgeneration mismatch.

142

M.R. Aghamohammadi, A. Shahmohammadi / Electrical Power and Energy Systems 35 (2012) 138147

Start
Pre processing network structure
Adopt number of islands #Nisland Set scenario No. S=1

Search Islanding scenario #S Select initial points for Island search

Set Initial status of all buses to 0, indicating their availability in the search space for occupation by Ants.

Ants movement for finding islands

Is there any unoccupied bus connected to Ants K (1, 2 , )? Randomly select an unoccupied bus #J connected to Ant K(1, 2, ..)

Next scenario S=S+1

End of search for Ant K (Island K is formed )

Move Ant K toward bus #J and change bus status to K(1, 2, ..) Is Load-Generation balance satisfied?

Is Line over loading constraint satisfied?

Establish islanding scenario #S

Is more Islanding scenario needed?

End of Process

Fig. 1. Flow chart of the proposed approach for island identication.

11. Whenever an island is established, Eqs. (1) and (2) of the criterion 1 will be checked and if they are not satised the search process for all ants will be terminated. 12. For islanding scenarios which fullled criterion 1, criterion 2 will be checked using Eqs. (3)(7) and those scenarios satisfying this criterion constitute initial islanding scenarios for further examination. The proposed algorithm has been coded in Matlab.

6. Simulation studies In order to demonstrate the effectiveness and ability of the proposed algorithm for nding islanding scenarios, IEEE 39-bus test system is adopted for simulation studies. The proposed algorithm is applied for a critical operating condition with total load of 7613 MW in which unit G5 at bus 34 and line 25 are tripped due to disturbances and unit G7 at bus 36 has decreased its output by 300 MW. As it is mentioned in the previous section, the number

M.R. Aghamohammadi, A. Shahmohammadi / Electrical Power and Energy Systems 35 (2012) 138147

143

Fig. 2. IEEE 39-bus system splitting for 8th islanding scenario.

of islands should be decided based on the operational and dynamic characteristic of power network. In this respect, prior information about potential coherent groups of generator would be helpful in adopting proper number of islands. In this study, based on the number of potential coherent groups of generators, it is decided to separate the entire network into four islands. It is worth noting that improper number of islands may affect both search process and implementation of islanding scenario. Higher number of islands may decrease search speed and make island implementation more difcult. On the other hand, improper less number may cause coherent groups of generators to interfere in each other. For each generator, 20% of its nominal capacity is considered as a controllable reserve and 20% of each load is allocated for load shedding. The algorithm has been run on a PC with 2.26 GHZ CPU, 2 GB RAM and 3 MB Cache with an average calculation time about 1 s for nding each islanding scenario. For the given operating point, 10 islanding schemes have been found. Fig. 2 shows system splitting conguration for the 8th islanding scenario. Table 1 shows generation, load, generation reserve and load shedding capacities in each island of the 8th islanding scenario. In this table, also the amounts of active power imbalance, applied generation control (increase/decrease) and applied load shedding for each island are shown. Table 2 shows boundary lines between each two islands with total power transfer between them before splitting. In fact, in order to establish islands these lines must be tripped. Tables 36 show the detailed amount of load shedding and generation rescheduling carried out in four established islands of the 8th islanding scenario. Also Tables 710 show lines power ow in each island after establishment of the island and applying control actions. As it can be seen in each island, loadgeneration balance and line overloading constraints have been satised by means of generation control and proper load shedding. Table 11 shows 10 islanding scenarios found by the proposed algorithm for the given operating condition. It is clear that in order

to nd the most applicable scenario, other criteria mentioned in Section 3 should be checked and fullled. For 10 islanding scenarios, totally 40 islands have been identied which can be categorized in nine types as shown in Table 12. For example, island type 1 with 7 buses and 984 MW generations including generators No. 5 and 9 has been identied in all scenarios. Scenarios 1, 2, 4, 7 and 9 have identical types of islands. In other words, in spite of random initialization of the search algorithm and its probabilistic mechanism, the identied scenarios and island types are relatively similar indicating a convergence in the identication trend of the algorithm. After identifying the best islanding scenario, the most signicant step is its implementation into power system. It is clear that each islanding scenario will be established by tripping boundary lines between islands. However, tripping each boundary line carrying power can impose a transient and dynamic effect on the island and the remaining part of the system. For this purpose, a suitable algorithm should be developed to manage the order and priority of tripping boundary lines between islands.

7. Conclusion In this paper, a new algorithm based on a probabilistic search mechanism denoted as ant search algorithm has been proposed for identifying feasible islands in a controlled splitting strategy. The proposed algorithm is based on a probabilistic search mechanism which starts from a number of randomly selected initial points. This algorithm is principally able to split the network into a predened number of controlled islands. However, the proper and feasible number of islands should be decided based on the dynamic and operational characteristic of power system at the moment of islanding. In this approach, with respect to available generation control reserve and load shedding capacity, the ability of islands for preserving loadgeneration balance and avoiding line overloading are evaluated as criteria for island establishment. The

144

M.R. Aghamohammadi, A. Shahmohammadi / Electrical Power and Energy Systems 35 (2012) 138147

Table 1 Generation, generation reserve, load and load shedding in the islands of the 8th islanding scenario. Island No. 1 2 3 4 No. of buses 11 7 10 11 PG (MW) 1848 984 1995 1854 PLoad (MW) 2421 1103 2230 1859 PResev (MW) 185 98 199 185 PShed (MW) 484 221 446 372 P. imbalance (MW) 573 119 236 5 Dec. PG (MW) 0 0 0 0 Increase PG + Pshed (MW) 573 1196 236 5

Table 2 Transmission lines connecting islands (with their total transfer power in MW) prior to splitting in the 8th islanding scenario. Island No. 1 2 3 4 No. of buses 11 7 10 11 Island 1 0 0 171736 (144) Island 2 0 21 (319) 30 (84) Island 3 0 21 (319) 20 (384) Island 4 171736 (144) 30 (84) 20 (384)

Table 3 Loadgeneration, generation control and load shedding in island 1 from 8th islanding scenario. Bus Before splitting (MW) Load 1 39 9 8 5 6 11 10 32 12 7 Total 0 1339 0 633 0 0 0 0 0 165 284 2421 Gen. 0 1061 0 0 0 0 0 0 787 0 0 1848 After splitting(MW) Load 0 1114 0 533 0 0 0 0 0 146 239 2033 Gen. 0 1167 0 0 0 0 0 0 866 0 0 2033 0 225 0 100 0 0 0 0 0 19 45 388 0 106 0 0 0 0 0 0 79 0 0 185 Load shed. (MW)

DPG (MW)

Table 4 Loadgeneration, generation control and load shedding in island 2 from 8th islanding scenario. Bus Before splitting (MW) Load 27 17 34 26 29 28 38 Total 341 0 0 169 344 250 0 1103 Gen. 0 0 0 0 0 0 984 984 After splitting (MW) Load 337 0 0 164 337 245 0 1082 Gen. 0 0 0 0 0 0 1082 1082 4 0 0 5 7 5 0 21 0 0 0 0 0 0 98 98 Load shed. (MW)

DPG (MW)

Table 5 Loadgeneration, generation control and load shedding in island 3 from 8th islanding scenario. Bus Before splitting (MW) Load 24 16 19 20 33 23 21 22 36 35 Total 374 399 0 825 0 300 332 0 0 0 2230 Gen. 0 0 0 0 787 0 0 0 398 810 1995 After splitting (MW) Load 370 394 0 810 0 294 327 0 0 0 2194 Gen. 0 0 0 0 866 0 0 0 437 891 2194 5 5 0 15 0 7 5 0 0 0 36 0 0 0 0 79 0 0 0 40 81 199 Load shed. (MW)

DPG (MW)

M.R. Aghamohammadi, A. Shahmohammadi / Electrical Power and Energy Systems 35 (2012) 138147 Table 6 Loadgeneration, generation control and load shedding in island 4 from 8th islanding scenario. Bus Before splitting(MW) Load 13 14 4 3 2 30 25 15 18 31 37 Total 0 0 607 391 0 0 272 388 192 10 0 1859 Gen. 0 0 0 0 0 466 0 0 0 715 673 1854 After splitting(MW) Load 0 0 607 391 0 0 272 388 192 10 0 1859 Gen. 0 0 0 0 0 448 0 0 0 746 666 1859 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 19 0 0 0 31 7 5 Load shed (MW)

145

DPG (MW)

Table 7 Parameters of transmission lines in island 1 from 8th islanding scenario. Line number 2 9 10 11 12 13 14 15 16 35 38 From bus 1 5 5 6 6 7 8 9 10 12 10 To bus 39 6 8 7 11 8 9 39 11 11 32 R (pu) 0.001 0.0002 0.0008 0.0006 0.0007 0.0004 0.0023 0.001 0.0004 0.0016 0 X (pu) 0.025 0.0026 0.0112 0.0092 0.0082 0.0046 0.0363 0.025 0.0043 0.0435 0.02 1/2B (pu) 0.375 0.0217 0.0738 0.0565 0.06945 0.039 0.1902 0.6 0.03645 0 0 Power ow (MW) 0.00 319.94 319.94 399.44 719.38 160.92 52.38 52.38 865.85 146.47 865.85 Pmax (MW) 800 550 1100 1100 1200 600 550 550 1500 550 1800

Table 8 Parameters of transmission lines in island 2 from 8th islanding scenario. Line number 26 31 32 33 34 40 45 From bus 17 26 26 26 28 17 29 To bus 27 27 28 29 29 34 38 R (pu) 0.0013 0.0014 0.0043 0.0057 0.0014 0.0009 0.0008 X (pu) 0.0173 0.0147 0.0474 0.0625 0.0151 0.018 0.0156 1/2B (pu) 0.1608 0.1198 0.3901 0.5145 0.1245 0 0 Power ow (MW) 0.00 337.01 220.80 279.89 465.39 0.00 1082.44 Pmax (MW) 500 900 600 700 1100 1500 2300

Table 9 Parameters of transmission lines in island 3 from 8th islanding scenario. Line number 22 23 24 27 28 29 39 41 42 46 From bus 16 16 16 21 22 23 19 22 23 19 To bus 19 21 24 22 23 24 33 35 36 20 R (pu) 0.0016 0.0008 0.0003 0.0008 0.0006 0.0022 0.0007 0 0.0005 0.0007 X (pu) 0.0195 0.0135 0.0059 0.014 0.0096 0.035 0.0142 0.0143 0.0272 0.0138 1/2B (pu) 0.152 0.1274 0.034 0.12825 0.0923 0.1805 0 0 0 0 Power ow (MW) 56.56 353.38 15.79 680.85 209.94 353.73 866.10 890.79 437.43 809.54 Pmax (MW) 500 1100 550 1700 550 1200 1800 2000 1700 2100

nature of the proposed search mechanism is very simple and almost independent of network size; therefore, it is able to implement any further static and dynamic constraints in the process of search algorithm. The proposed approach is able to identify a variety of islanding scenarios satisfying basic criteria. Therefore, provision of variety of islanding scenarios enables operator to effectively implement his engineering understanding and experience in the process of island identication. The proposed approach is so exi-

ble for implementing further constraints and system characteristic in the process of island identication. Reactive power loadgeneration balance and minimum number of boundary lines between islands are examples of additional constraints which can be handled into the search process. The proper number of islands to be established is mainly decided with respect to coherent group of generators. The proposed algorithm can be carried out many times to identify several islanding scenarios. However, the simulation

146

M.R. Aghamohammadi, A. Shahmohammadi / Electrical Power and Energy Systems 35 (2012) 138147

Table 10 Parameters of transmission lines in island 4 from 8th islanding scenario. Line number 3 4 5 6 8 18 19 37 43 44 From bus 2 2 3 3 4 13 14 4 25 2 To bus 3 25 4 18 14 14 15 31 37 30 R (pu) 0.0013 0.007 0.0013 0.0011 0.0008 0.0009 0.0018 0 0.0006 0 X (pu) 0.0151 0.0086 0.0213 0.0133 0.0129 0.0101 0.0217 0.025 0.0232 0.0181 1/2B (pu) 0.1286 0.073 0.1107 0.1069 0.0691 0.08615 0.183 0 0 0 Power ow (MW) 841.61 394.06 259.40 191.61 388.10 0.00 388.10 735.70 665.72 447.54 Pmax (MW) 1100 850 600 500 550 550 500 1700 1700 1600

Table 11 10 Islanding scenarios. Scenario 1 Island number 1 2 3 4 No. of buses 7 11 13 8 39 2 1 2 3 4 10 7 12 10 39 3 1 2 3 4 12 5 11 11 39 4 1 2 3 4 5 12 10 12 39 5 1 2 3 4 10 11 11 7 39 6 1 2 3 4 8 15 7 9 PG (MW) 984 1995 1502 2200 6681 2200 984 1502 1995 6681 1995 984 1854 1848 6681 984 1502 2200 1995 6681 787 2914 1995 985 6681 2200 2710 984 787 6681 7 1 2 3 4 10 7 12 10 39 8 1 2 3 4 11 7 10 11 39 9 1 2 3 4 5 12 9 13 39 10 1 2 3 4 9 10 13 7 39 2200 984 1502 1995 6681 1848 984 1995 1854 6681 984 1502 2200 1995 6681 787 1995 2914 985 6681 PL (MW) 1103 2618 1699 2193 7613 2826 1103 1454 2230 7613 2230 1103 1859 2421 7613 1103 1454 2826 2230 7613 1082 2810 2618 1103 7613 2193 3235 1103 1082 7613 2826 1103 1454 2230 7613 2421 1103 2230 1859 7631 1103 1699 2193 2618 7613 837 2230 3443 1103 7613 Gen. decrease 0 0 0 7 7 0 0 48 0 48 0 0 0 0 0 0 48 0 0 48 0 105 0 0 105 7 0 0 0 7 0 0 48 0 48 0 0 0 0 0 0 0 7 0 7 0 0 0 0 0 Gen. increase + L. shed. 119 623 197 0 939 626 119 0 235 980 235 119 5 573 933 119 0 626 235 989 295 0 624 118 1029 0 525 119 295 939 626 119 0 235 980 573 119 236 5 933 119 197 0 623 939 50 236 528 118 932

M.R. Aghamohammadi, A. Shahmohammadi / Electrical Power and Energy Systems 35 (2012) 138147 Table 12 Types of identied islands in 10 islanding scenarios. Type Buses PG (MW) 1 7 984 10 2 11 1995 9 3 9 2200 6 4 12 1502 5 5 9 787 3 6 11 1854 2 7 11 1845 2 8 12 2914 2 9 15 2710 1

147

results show that in spite of random initialization of search process for each islanding scenario, often denite number of feasible islanding scenario will be found. The proposed algorithm has been applied on IEEE 39-bus network with four intended islands and simulation results clearly demonstrate its ability and efciency for identifying islanding scenarios. References
[1] Ford JJ, Bevrani H, Ledwich G. Adaptive load shedding and regional protection. Elect Power Energy Syst 2009;31:6118. [2] Rosas-Casals Marti, Sole Ricard. Analysis of major failures in Europes power grid. Elect Power Energy Syst 2011;33:8058. [3] Fu Xu, Wang Xifan. Determination of load shedding to provide voltage stability. Elect Power Energy Syst 2011;33:51521. [4] Maharana Manoj Kumar, Swarup KS. Graph theoretic approach for preventive control of power system. Elect Power Energy Syst 2010;32:25461. [5] Zhao Qianchuan, Sun Kai, Zheng Da-Zhong, Ma Jin, Lu Qiang. A study of system splitting strategies for island operation of power system: a two phase method based on OBDDs. IEEE Trans Power Syst 2003;18:155665. [6] Sun Kai, Sidhu Tarlochan S, Jin Ming. Online pre-analysis and real-time matching for controlled splitting of large-scale power networks. Proc Int Conf Future Power Syst 2005:16. [7] Taylor CW, Erickson DC. Recording and analyzing the July 2 cascading outage [Western USA power system]. IEEE Comput Appl Power 1997;10:2630.

[8] Thaden M. Analysis of a major load island outage on the POTOMAC electric power company system. IEEE Trans Power Syst 1999;14:30611. [9] Vieira X et al. The March 11th 1999 blackout: short-term measure to improve system security and overview of the reports prepared by the international experts CIGR session. SC 39 workshop on large disturbances, Paris; 2000. [10] Jonsson M. Line protection and power system collapse, Chalmers Univ. Technology, School of Electrical and Computer Engineering, Gteborg, Sweden, 393L; 2001. [11] Thapar J, Vittal V, Kliemann W, Fouad AA. Application of the normal form of vector elds to predict interarea separation in power systems. IEEE Trans Power Syst 1997;12:84450. [12] Vittal V, Kliemann W, Ni Y-X, Chapman DG, Silk AD, Sobajic DJ. Determination of generator groupings for a islanding scheme in the Manitoba Hydro system using the method of normal forms. IEEE Trans Power Syst 1998;13:134551. [13] Agematsu S, Imai S, Tsukui R, Watanabe H, Nakamura T, Matsushima T. Islanding protection system with active and reactive power balancing control for Tokyo metropolitan power system and actual operational experiences. In: Proc inst elect eng seventh int conf develop power syst protection; 2001. p. 3514. [14] Archer BA, Davies JB. System islanding considerations for improving power system restoration at Manitoba Hydro. Proc IEEE Can Conf Elect Comput Eng 2002;1:605. [15] Tsai M-S. Development of islanding early warning mechanism for power systems. In: IEEE power eng soc summer meeting, vol. 1; 2000. p. 226. [16] Sun K, Zheng D, Lu Q. Splitting strategies for islanding operation of large-scale power systems using OBDD-based methods. IEEE Trans Power Syst 2003;18:912. [17] Sun K, Zheng D, Lu Q. Searching for feasible splitting strategies of controlled system islanding. In: IEE proceeding online no.20050168; 2006. [18] Ahmed SS, Sarker NC, khairuddin AB, et al. A scheme for controlled islanding to prevent subsequent blackout. IEEE Trans Power Syst 2003;18:13643. [19] Seca L, Lopes JAP. Intentional islanding for reliability improvement in distribution networks with high DG penetration. In: Future power system, international conference, Amsterdam; 2005. p. 05. [20] Sun K, Zheng D, Lu Q. A simulation study of OBDD-based proper splitting strategies for power system under consideration of transient stability. IEEE Trans Power Syst 2005;20:38999.