نویسندگان
^{1} شرکت برق منطقه ای اصفهان، اصفهان، ایران
^{2} گروه برق  قدرت، دانشکده فنی و مهندسی، دانشگاه اصفهان، اصفهان، ایران
چکیده
کلیدواژهها
موضوعات
عنوان مقاله [English]
نویسندگان [English]
Installing new energy sources as redundant blackstart (BS) units is an efficient way to enhance the speed of power system restoration, especially when there is a high risk that the available power plants considered as BS units fail to operate. In this regard, this paper provides a new optimal design for the placement of the Gas Turbine (GT) as the redundant energy source to improve the power system performance during both restoration and normal conditions. In doing so, there will be contradictory objective functions to be minimized. Therefore, a multiobjective problem (MOP), as a mixed integer linear programming (MILP), is defined. The Pareto optimal solutions of the MOP are obtained by using a new populationbased metaheuristic technique, called Crow Search Algorithm (CSA). Two power systems are used for the validation of the proposed method. The simulation results show that the system can benefit from this method not only to increase the capability of blackstart generation, but also to improve the power system performance in normal conditions. During the restoration process, it also provides the optimal startup sequences of nonblackstart (NBS) units with the optimal transmission paths.
کلیدواژهها [English]
1. Introduction [1]
The modern power systems are being controlled in a high level of reliability and security. However, different unpredicted events may occur and, in turn, cause to have a wide area of cascading failures and then a blackout for the entire system [12]. This can impose huge economic losses and negative social impacts. Therefore, the quick restoration of a power system after a blackout is one of the most important concerns for power system dispatchers, operators and planners [3].
The process of power system restoration includes three stages: preparation, transmission paths reconfiguration and load restoration [4]. In the preparation stage, the blackstart units are commissioned, and then transmission paths are connected to crank nonblackstart units (NBSU) in order to be started up [5]. The objective of this stage is mainly to maximize the generation capability. In this regard, many researchers have formulated this objective as an optimization problem [36]. In [3] and [5] authors propose an algorithm that formulates generator startup sequence as a MILP problem. In [3] the firefly algorithm (FA) is used to search the final optimal sequence of NBSUs, transmission paths and load pick up sequence by the aid of renewable energy sources (RES) already installed in power system. In [5] an optimal generator startup strategy is addressed to find a sequence of starting time for all NBSUs and also to update the generation capability during system restoration. In [6], an initial restoration sequence of generators is obtained by using a mixed integer linear problem (MILP) method. Then, the final restoration sequence is manually acquired by using this initial guess after many trials. A decision making support system is addressed in [78] to search the optimal blackstart strategy by using a concept of the vaguevalued fuzzy measure. In [9] a steady state model is developed to optimize the total generator power output through searching the optimal startup sequence of NBSUs by using the shortest path decision. In [10], the optimal startup sequences of generators are designed by applying an algorithm based on Dijkstra method for finding the shortest path during power system restoration. In [11], the optimal network reconfiguration was designed into two separately steps. In the first step, the optimal sequence design of generating units was done to maximize the restored generation capability, and in the second step, the optimal transmission path is obtained to crank the generating units. In [12] a combined design of the generating startup sequence and the transmission path search is presented as an optimization problem and solved by using firefly algorithm.
In all researches published so far, to the best of our knowledge, the selection of blackstart unit (BSU) as a starting point of restoration process has been based on the existed units within the power system without performing any design procedure to locate these units. In this regard, although the hydro and Gas Turbine (GT) power plants that are generally considered as the BSUs, due to the high speed of commissioning, it will be possible that these plants do not have enough capabilities to crank NBS units [13]. Some of the hydro power plants are not available because of the insufficient flow of the water or the low level of the stored water behind the dam [14]. Moreover, when the alreadyinstalled gas turbine power plants continuously work in the system, there is a high risk that after blackout, the technical problem in fuel system causes to trip the unit or it may not permit to start the unit for restoration preparation [15]. Furthermore, in both gas turbine and hydro power plants, the failure of control systems may stop the process of starting BSU [1516]. Therefore, considering redundant BSUs is necessary to crank and to increase the speed of restoration [1719].
There are many situations during the normal condition of a power system that the system should be able to respond to the sudden load demands very quickly. In these cases the new installed GTs that can be started very fast could be one of the best options to help the system increase the generations in the shortest possible time [2021]. Therefore, it will be more economical to place these units in the locations where the minimum power losses are achieved. They must also justify another important constraint which is the voltage violation during the normal condition. Therefore, the problem of the optimal placement of GTs, as the good candidates for redundant BSUs, will be solved by increasing the capability of blackstart generation and also improving the power system performance in normal conditions. In doing so, a multiobjective function is used to obtain the best locations of GTs, the optimal transmission path and the optimal generators startup sequences by considering the blackstart capability criterion.
There are several techniques for finding the optimal transmission path in a power system including expert system [22], artificial neural networks [23], power transfer distribution factor (PTDF)based approach [24], linear programmingbased approach [25], heuristic approaches [26] and Dijkstra algorithm [27]. Among all these techniques, Dijkstra algorithm has a simpler structure and, therefore, the computational complexity of any proposed design is decreased by using this algorithm [28]. Also, the CPU time of running the proposed optimization model during the optimization process is decreased.
The multiobjective function consists of minimizing the Unavailable Energy Capability (UEC) of the system during the restoration process and also minimizing the voltage deviations of buses and active power losses in transmission system during normal condition to achieve the best Pareto optimal set. Then, a new populationbased metaheuristic technique, called crow search algorithm (CSA) based on fuzzy decision making approach is used to obtain the Pareto optimal solutions. The idea of using CSA is initiated from one of the real and intelligent behavior of storing, hiding and retrieving process of crow’s food by themselves [29]. The high level of efficiency, accuracy and robustness to search the Pareto optimal solution and the finding capability of the proposed algorithm make it be a desirable approach for optimizing the MOP addressed in this paper. The main contributions of this paper are as follows:
The 39bus New England power system and a 32bus power system are employed as two test systems. The simulation results show that not only the proposed method is capable of finding the best locations of the redundant GTs, but also it finds the optimal startup sequences of NBS units with the optimal transmission paths in order to speed up the restoration process. It also improves the performance of the power system in normal conditions.
The rest of the paper is organized as follows: Section 2 describes the problem formulation. In Sections 3 and 4, the solution approach (MOCSA) is given to solve the proposed multiobjective problem. Simulation results are given in Section 5 and finally, Section 6 concludes the paper.
2. Problem Formulation
The formulation of the multiobjective model for searching the location of a GT including three objective functions and their constraints is given in this section.
2.1. Objective Functions
The employed objective functions for the proposed multiobjective model are explained in this subsection.
2.1.1. Unavailable Energy Capability Objective Function
The first objective function is used to minimize the total generation capacity (MWh) which is not available after blackout [3] and is given by the following equation [3]:
(1) 
The optimal startup sequence of NBS units and the optimal locations of GTs are obtained by the optimization of the first objective function. The optimal transmission paths are also determined when the objective function f_{1}is calculated. This will be described in the next subsection.
2.1.1.1. Optimal Path Search Algorithm
A minimum spanning tree method is used here to find the shortest paths between two nodes in network and can be shown as a graph [30]. This algorithm, which is an iterative process, has been introduced by W. Dijkstra [30]. A simple example of this algorithm is shown in Fig. 1. In this figure, a shortest path is found between node A (source node) and node E (destination node). The start point of this algorithm is shown by coloured node A (step 1). In the first iteration, the algorithm searches the adjacent node from the source node, which must be the nearest node to the source node (step 2). In the second iteration, the algorithm searches the second nearest node from the source node
(step 3). This node must be the nearest node from the first node which was the closest to the source node. This procedure iteratively continues so that the minimum spanning tree for the source node is found (step 4). This tree gives an optimal path which includes all nodes.
Fig (1): Dijkstra algorithm [30]
2.1.2. Minimization of Active Power Loss
Minimizing the active power loss is formulated as follows [31]:
(2) 
2.1.3. Voltage Violation
The following objective function is given to reduce the voltage violation in all PQ buses [31]:
(3) 
2.2. Decision Variables
The decision variables are optimally obtained after completing the optimization process in order to achieve the best optimal solution. In the proposed design method, a constrained optimization problem will be described as a multiobjective function and minimized subject to and . The and are defined as the decision variables and imply all startup times of NBS units and the location of a GT (redundant BSU) respectively.
2.3. Multiobjective Design
In multiobjective design it is impossible to obtain an optimal solution when all objective functions are simultaneously being optimized, because the objective functions are different and each function is not compromised with other functions. In this condition, a decision maker chooses a “most effective” answer as an optimal solution [32]. Generally, the equation of a MOP can be formulated as follows;
Optimize
(4) 
The vector is defined as decision variables and obtained by solving Eqn. (4). Also, this MOP is executed subject to regard the total number of all constraints. These constraints consist of inequality () and equality () constraints as given in Eqns. (5) and (6), respectively.
(5) 

(6) 
In this study, Eqns. (5) and (6) include all equality and inequality constraints given in Eqns. (1), (2) and (3) which are the starting times of NBSUs and the generation capability of BSUs, active and reactive power balance and also the voltages of buses.
Generally, the proposed MOP is optimized by storing a group of most desirable answers in a repository and renewing the group at any iteration [33]. In this technique, the best acquired answers are defined as nondominated answers or Pareto optimal set. An answer can be determined as a Pareto optimal set if and only if Eqn. (7) is satisfied.
(7) 
In Eqn. (7) is defined as the Pareto optimal set of the possible search space of. It also implies that enhancing one objective function by an allowable answer will be gained by degrading the optimal solution of at least one of the other objective functions.
2.4. Best Acceptable Solution
In the optimization process, it is required that a solution be considered between all nondominated answers of the Pareto optimal set as the best allowable answer [33]. In this paper, a fuzzy mechanism is used for three objective functions (f_{1}, f_{2}, f_{3}) because their values are not precise. In this regard, a fuzzy membership function is defined for achieving the most desirable answer. For all individual recorded in repository, the following equation is formulated [33]:
(8) 
where and are described as the maximum and minimum limits of each objective function respectively. These amounts are calculated by comparing with the values acquired by the single optimization of each objective function.
The normalization of the membership of the recorded individuals could be done according to the following equation [34]:

(9) 
where ,, are the total number of objective functions, the number of Pareto optimal answers in repository and the weight of K_{th} objective function respectively. Finally, the best compromise solution is considered as the solution which has the maximum .
2.5. Repository Size Control
One of the main factors in the design of any evolutionary optimization approach is to control its repository’s size [3536]. In this paper, a fuzzy clustering technique is used to control the size of the repository without changing its characteristics [35]. In doing so, each solution of the repository is defined as a cluster with definite radius. Neighbor clusters are combined until the proper repository size is achieved. In the combining stage, the member with the higher membership value of combined clusters is chosen to be recorded in the repository. The flowchart of the whole mechanism using fuzzy based clustering is given in Appendix.
3. Crow Search Algorithm
The idea of crow search algorithm (CSA) was firstly created from one of the real and intelligent behavior of crows by A. Askarzadeh in 2016 [29]. This algorithm formulates the storing, hiding and retrieving process of crow’s food by themselves. The full details of CSA are given in [29].
One of the most important specifications of metaheuristic algorithms is satisfying a good balance between diversification and intensification [37]. In CSA, intensification and diversification are principally balanced by the parameter of awareness probability (AP). By decreasing the value of AP, the searching process will go into a local area in CSA where a current good solution is obtained in this area. On the other hand, increasing the value of AP conducts searching process in CSA into the global area (randomization). Therefore, diversification is increased by applying the large values of AP.
4. Implementation of CSA in Multiobjective Design for Optimal Placement of GTs
Now, the design procedure for implementation of CSA in multiobjective design is described as follows:
Step 1: Determine the required data which are flock size of crows (N), flight length (fl), awareness probability (AP), maximum iteration, and Pareto archive dimension.
Step 2: Generate the initial random population of crow’s position and their memories.
Step 3: Compute the value of objective functions for each crow by solving Equation 4.
Step 4: Investigate the constraints of the problem by using Eqns. (1), (2) and (3). If the constraints are not satisfied, consider a high value of penalty for objective functions related to the current crow, otherwise, ignore it.
Step 5: Search the nondominated solutions among the possible solutions and record them in the Pareto repository according to Eqns. (8) and (9).
Step 6: Check the repository size. If the size of repository is more than the maximum predefined size, remove the extra member with fuzzy based clustering mechanism.
Step 7: Generate the new population of crow’s position by using Eqns. (10) and (11).
Step 8: Check the feasibility of the new population of crow’s position. If the new position is possible, go to step 9. Otherwise, go to step 7.
Step 9: Compute the value of objective functions for each crow by solving Equation 4.
Step 10: Investigate the constraints of the problem by using Eqns. (1), (2) and (3). If the constraints are not satisfied, consider a high value of penalty for objective functions related to the current crow, otherwise, ignore it.
Step 11: Update the repository by replacing the new nondominated solutions according to Eqns. (8) and (9).
Step 12: Check the repository size. If the size of repository is more than maximum predefined size, remove the extra member with fuzzy based clustering mechanism.
Step 13: Check the convergence criterion. If the maximum iteration number is reached, stop the execution of the algorithm. Otherwise go back to step 7.
The flowchart of the abovementioned steps is given in Fig. 2.
5. Simulation Results
In this section, the proposed method is performed on two different power systems to show its efficient performance during restoration and normal conditions.
5.1. The 39bus New England Power System
The single line diagram of the 39bus New England power system is depicted in Fig. 3 and its information can be found in [38]. This system is segmented into two subsystems only at restoration condition according to [3], and each subsystem must have at least one blackstart unit [39]. Therefore, the number of already installed BSUs is assumed to be 2, which are G6 and G10 [3, 5]. As mentioned in section 1, it is essential that the redundant BSUs (GTs) be considered for each power system, and their optimal locations are obtained.
Fig (2): Flowchart of application of CSA on multiobjective design
Fig (3): IEEE 39bus power system [37]
Now, the proposed design procedure is performed in two stages. First, the single objective design is done to specify and to use the limit values of three objective functions f_{1}, f_{2}and f_{3} ((Eqns. (1), (2) and (3)) for the second stage (multiobjective design). Then, the multiobjective design is carried out to determine the optimal solution (or optimal decision variables), which are the startup times of NBS units and the locations of GTs. The required data of generators for blackstart process are given in Table 1 according to [5]. In this table, refers to the cranking time for generator to start to ramp up for connecting with system, and is the rate of generator ramping.
Table (1): Characteristics of generating units in 39bus system [4]
unit 
Type 
P_{max }(MW) 
P_{start} (MW) 
R_{r} (MW/hr) 
T_{max} (min) 
T_{min} (min) 
G1 
Coal 
572.9 
5.5 
215 
N/A 
40. 
G2 
Nuc. 
650 
8 
246 
N/A 
N/A 
G3 
Nuc. 
632 
7 
236 
120 
N/A 
G4 
Coal 
508 
5 
198 
N/A 
70 
G5 
Coal 
650 
8 
244 
60 
N/A 
G6 
Nuc. 
560 
0 
214 
N/A 
N/A 
G7 
Coal 
540 
6 
210 
N/A 
N/A 
G8 
Nuc. 
830 
13.2 
346 
N/A 
N/A 
G9 
Nuc 
1000 
15 
384 
N/A 
N/A 
G10 
Hyd. 
250 
0 
162 
N/A 
N/A 
5.1.1. Single Objective Design
In this part, single objective design is performed by using CSA algorithm. The extracted data of this part are used as the limits for the multiobjective approach.
5.1.1.1. Restoration Condition
In this subsection, the objective function f1 (Eqn. (1)) is minimized by using CSA so that all optimal startup times of NBS units and the best locations of GTs (redundant BSUs) are obtained and the results are given in Table 2. The convergence curve of f1 is also shown in Fig. 4. The last row of Table 2 shows the best value of the objective function f1 used for the limits considered in multiobjective method. By using the optimal search path algorithm explained in section 2.1.1.1, the optimal startup sequences of NBSU units are obtained and given in Table 2. In this regard, each subsystem includes one redundant BSU and four NBSUs. . It should be mentioned that the capacity of each gas turbine is selected to be 50 MW because the total amount of power needed for starting all NBS units (Pstart) in each subsystem is 50 MW.
Fig (4): CSAbased convergence curve of UEC
Table (2): Optimal design of generators startup process using CSAbased design considering redundant BSU
Subsystem 1 
Gen 
G10 
G1 
G2 
G8 
G9 
BSU 
BSU 
 
 
 
 
 
√ 

NBSU 
√ 
√ 
√ 
√ 
√ 
 

25 
40 
40 
30 
40 
15 

Location of BSU 
 
 
 
 
 
Bus 3 

Subsystem 2 
Gen 
G4 
G3 
G5 
G6 
G7 
BSU 
BSU 
 
 
 
 
 
√ 

NBSU 
√ 
√ 
√ 
√ 
√ 
 

70 
40 
30 
30 
35 
15 

Location of BSU 
 
 
 
 
 
Bus 16 

Value of UEC 
2334.9 
The colored optimal path of cranking power for all NBSUs is depicted in Fig. 5. In this figure, the sequence of startup for all NBSUs is determined by a specific color. For more clarification, the details of these paths are given in Table 3.
Fig (5): Optimal generator startup of IEEE 39 bus power system using redundant BSU
As mentioned before, G6 and G10 have been considered as main BSUs. A comparison is performed between two different cases where the system is equipped with new GTs as redundant BSUs and the case where new GTs are not installed. The results are presented in Table 4. The results given in this table show that the value of UEC for all cases is desirably decreased when new GTs are used when either G6 or G10 fails to operate. For example, in case 1 when G6 fails to operate, the value of UEC by using new GTs (2309.9 min. p.u.) is lower than the ones without new GTs (2870.60 min. p.u.). For the worst situation (case 3) when both G6 and G10 fail to operate, by using new GTs as redundant BS units, the value of UEC is 2334.90 min. p.u., which means that the blackstart process is successfully performed.
Table (3): The optimal cranking power paths of NBSUs in New England power system
Location of redundant BSU 
NBS unit 
optimal cranking power paths 
Bus 3 
G1 
Bus: 3→2→1→39 
Bus 3 
G2 
Bus: 3→4→6→31 
Bus 3 
G8 
Bus: 3→2→25→37 
Bus 3 
G9 
Bus: 3→2→25→26→29→38 
Bus 3 
G10 
Bus: 3→2→30 
Bus 16 
G3 
Bus: 16→15→14→13→10→32 
Bus 16 
G4 
Bus: 16→19→33 
Bus 16 
G5 
Bus: 16→19→20→34 
Bus 16 
G6 
Bus: 16→21→22→35 
Bus 16 
G7 
Bus: 16→21→22→23→36 
Table (4): Effect of GTs as the redundant BSUs in blackstart process during different cases
Gen. unit 
1 
2 
3 
4 
5 
6 
7 
8 
9 
10 

With GTs 
t_{start} 
40 
40 
40 
70 
30 
30 
35 
30 
40 
15 

value of UEC 
2309.90 

Without GTs 
t_{start} 
40 
50 
50 
70 
55 
55 
60 
30 
40 
15 

value of UEC 
2870.60 

With GTs 
t_{start} 
40 
40 
40 
70 
45 
15 
30 
30 
40 
25 

value of UEC 
2320.5 

Without GTs 
t_{start} 
65 
60 
55 
70 
45 
15 
30 
55 
55 
55 

value of UEC 
3948.7 

With GTs 
t_{start} 
40 
40 
40 
70 
30 
30 
35 
30 
40 
25 

value of UEC 
2334.90 

Without GTs 
t_{start} 
∞ 
∞ 
∞ 
∞ 
∞ 
∞ 
∞ 
∞ 
∞ 
∞ 

value of UEC 
∞ 

5.1.1.2. Normal Condition
In this subsection, the values of objective functions f_{2}and f_{3} (Eqns. (2) and (3)) are separately minimized by executing the CSA based power flow analysis. Optimizing f_{2}and f_{3} gives the locations of GTs during the steady state condition. Figures 6 and 7 show the convergence curves of f_{2}and f_{3} respectively. The optimal values of these curves are shown in Table 5. The best locations of GTs and the minimum value of the related objective function are presented in this table. The third column of Table 5 gives different locations for GTs (bus 8 for f_{2} and bus 21 for f_{3}) because of having the conflict between the single objective functions during the optimization process. Also, the extracted data of active power loss (0.4164 p.u.) and voltage violation (2.2513 p.u.) given in this table are chosen as the limits for the multiobjective approach. As expected and given in Tables 2 and 5, the location of GTs are different because different objective functions have been used in different conditions.
Fig (6): CSAbased convergence curve of active power loss in steady state condition
Fig (7): CSAbased convergence curve of voltage violation in steady state condition
Table (5): Optimal design of objective function during CSAbased power flow analysis
Objective function 
Value of objective function 
Locations of New GTs 
Active power loss (p.u) 
0.4164 
Bus 8 
Voltage violation (p.u) 
2.2513 
Bus 21 
To show the efficiency of Dijkstra algorithm during searching the optimal cranking paths, a comparative analysis between this algorithm and heuristic algorithms is performed. The results shown in Table 6 imply that the CPU time of Dijkstra algorithm (1105 s) for finding the optimal paths is less than the heuristic algorithms including PSO (1240 s) and CSA (1235 s). Also, executing the program in 100 times, Dijkstra algorithm gives the best cranking paths in 95% of the total trials which is much more than the other algorithms; 87% and 85% for CSA and PSO algorithms respectively.
Table (6): Comparison of the results for different algorithms to obtain optimal path
Criteria 

Algorithm 


Dijkstrabased 
CSAbased 
PSObased 
CPU time (s) 
1105 
1235 
1240 
Number of iteration 
32 
43 
47 
No. global solution 
95 
87 
85 
5.1.1.3. Multiobjective Design
In this section, the CSA based multiobjective design is executed to attain the optimal decision variables (all optimal startup times of NBS units and the best locations of GTs) by the coordinated optimization of three objective function f_{1}, f_{2} and f_{3} ((Eqns. (1) , (2) and (3)), and results are shown in Table 7. The Pareto optimization curve is displayed in Fig. 8 in which the red star is assigned as the best compromise solution. According to Table 7, the optimal values of objective functions are more than the optimal ones obtained in single objective design (see Tables 2 and 5). This is because of the fact that all objective functions are simultaneously minimized in the multiobjective design. Table 6 also shows that the value of UEC (2522.37 p.u. min) obtained in MOP design is lower than the ones obtained in Table 4 for cases when new GTs are not used. This means that the multiobjective optimization method has a desirable performance in blackstart process. Also, the startup times of G2, G3, G5 and G7 in multiobjective design are less than the cases where new GTs are not used as given in Table 4.
Fig (8): Pareto curve of CSAbased multiobjective method
Table (7): Optimal multiobjective design using CSAbased optimization
Subsystem 1 
Gen 
G10 
G1 
G2 
G8 
G9 
redundant BSU 
BSU 
 
 
 
 
 
√ 

NBSU 
√ 
√ 
√ 
√ 
√ 
 

35 
40 
30 
35 
55 
15 

Location of BSU 
 
 
 
 
 
Bus 5 

Subsystem 2 
Gen 
G4 
G3 
G5 
G6 
G7 
redundant BSU 
BSU 
 
 
 
 
 
√ 

NBSU 
√ 
√ 
√ 
√ 
√ 
 

70 
45 
35 
35 
25 
15 

Location of BSU 
 
 
 
 
 
Bus 24 

Value of UEC (p.u . min) 
2522.37 

Active power loss (p.u) 
0.426 

Voltage violation (p.u) 
2.2570 
It should be noted that all objective functions are weighted by the same coefficient (w_{1}=w_{2}=w_{3}=0.33). w_{1}, w_{2}and w_{3} show the weights of three objective functions of UEC, active power loss and voltage violation, respectively. In order to show the reason of choosing these weights, two different solutions with different weights are used for IEEE 39bus power system and the results are shown in Table 8. As can be seen in this table, the value of UEC for solution 2 (2522.3 pu.m) is very close to solution 1 (only 0.8% difference). However, the values of active power loss and voltage violation in solution 2 (0.426 pu and 2.257 pu) have been decreased in comparison with solution 1 by 7% and 1.2 % respectively. This means that considering similar weight coefficients for objective functions in multiobjective design desirably decreases the active power loss and voltage violation in normal operation and also maintains the value of UEC close to solutions 1 and 2.
Table (8): Results of different solution of GT optimal placement for IEEE 39 bus power system
Solution 
Value of UEC (pu.min) 
Value of voltage violation (pu) 
Value of active power loss (pu) 
2500.5 
2.284 
0.456 

2522.3 
2.257 
0.426 
Another analysis is also performed for verifying the normal condition. The active power loss and the voltage violation of the system are computed before and after installation of new GTs. The results are given in Table 9. As can be seen in this table the active power loss and voltage violation are efficiently declined when new GTs are used. For example, the active power loss after using new GTs decreases about %7.66 for case 3 (max load). Also, the value of voltage violation decreases about %1.47.
To assess the performance of CSA algorithm, a comparison is also done between CSA, particle swarm optimization (PSO) and genetic algorithm (GA) methods for three cases of the best, average and worst solutions. The results of this comparative analysis are presented in Table 10. As shown in this table, the proposed CSA provides more desirable results in comparison with other two algorithms. For example, UEC amount in “best solution” part is 2522.37 for CSA method while this amount is 2528.45 for PSO and 2533.71 for GA. Executing the program in 100 times, CSA achieves to the best solution in 91% of the total trials which is much more than the other algorithms; 85% and 78% for PSO and GA algorithms respectively. This result shows the robustness of the proposed algorithm. Moreover, CPU time of CSA is lower than that of PSO because it is minimized to obtain the optimal solution in less iteration, needs fewer members, and it also avoids from being trapped in the local solution. The amount of the standard deviation is also lower for CSA than the others. As can also be seen in Table 10, the number of global solutions in this algorithm is higher than the others, and its standard deviation of has lower value in comparison with other algorithms.
Table (9): Extracted data during power flow analysis for normal condition
Objective function 
Active power loss 
Voltage violation 

Case 1 (min load) 
Without new GTs 
0.4214 
2.2513 
With new GTs 
0.4135 
2.2291 

Case 2 (mid load) 
Without new GTs 
0.4294 
2.2587 
With new GTs 
0.4182 
2.2320 

Case 3 (max load) 
Without new GTs 
0.4592 
2.2908 
With new GTs 
0.4265 
2.2570 
Table (10): Comparison of the results obtained by different algorithms considering multiobjective problem for 100 trials.
Solution 
Criteria 
Algorithm 

CSA 
PSO 
GA 

Best solution 
active power loss 
0.426 
0.429 
0.431 
voltage deviation 
2.257 
2.263 
2.264 

UEC 
2522.37 
2528.45 
2533.71 

0.0343 
0.0312 
0.0295 

Average solution 
active power loss 
0.4255 
0.433 
0.43 
voltage deviation 
2.375 
2.395 
2.44 

UEC 
2531.41 
2536.50 
2540.05 

0.0335 
0.0305 
0.0287 

Worst solution 
active power loss 
0.431 
0.437 
0.439 
voltage deviation 
2.45 
2.48 
2.55 

UEC 
2533.30 
2541.33 
2543.65 

0.0319 
0.03 
0.028 

SD 
0.000510 
0.000550 
0.00058 

No. global solution 
91 
85 
78 

CPU time (s) 
720.5 
735.2 
801.5 
5.2. Case2: 32bus Power System
In this subsection, a 32bus power system is considered to apply the proposed method. The required data and the single line diagram of this power system are given in Table 10 and Fig. 9, respectively. In this network, G4 and G5, which are already installed to supply electrical energy, are also considered as BS units. As reported in [17] there is a high risk that these two units fail during the restoration process. Therefore, the proposed multiobjective design method is applied to this system to find the location of the redundant BSU that should be installed in the system as well as determining the optimal starting times of NBSUs. Like 39bus power system this power system is also divided into two subsystems as shown in Table 11.
Fig (9): Single line diagram of 32bus power system
Table (11): Characteristics of generating units in 32bus power system
Generating unit 
G1 
G2 
G3 
G4 
G5 
G6 
Type of unit 
steam 
steam 
steam 
Gas 
Gas 
Gas 
P_{max} (MW) 
800 
800 
830 
780 
270 
420 
P_{start} (MW) 
12 
12 
14 
8 
5 
7 
R_{r} (MW/min) 
5 
5 
4 
10 
10 
7 
T_{max} (min) 
90 
90 
120 
N/A 
N/A 
N/A 
T_{min} (min) 
30 
30 
30 
N/A 
N/A 
N/A 
The results of this analysis are given in Table 12. The Pareto optimization curve is depicted in Fig. 10. In this figure, the best compromise solution is optimally obtained and shown by the magenta star considering. The value of w_{i} in Eqn. (9), as mentioned for 39bus power system, is also considered to be equal () for this system.
The optimal cranking power paths of NBSUs are given in Table 13 according to the obtained data of CSAbased multiobjective design. As can be seen in Tables 11 and 12, redundant BSUs (new GTs) are optimally located in buses 14 and 17 in subsystems 1 and 2 respectively. Since it is assumed that the maximum amount of power needed for starting all NBS units (P_{start}) in each subsystem is 50 MW, the capacity of each new GT is selected to be 50 MW.
Fig (10): Pareto curve of CSAbased multiobjective method
A power flow analysis is carried out to display the effective locations of new GTs in satisfying the operating constraints according to the proposed method. In this analysis, three different cases of minimum load, medium load and maximum load are chosen. The active power loss and the voltage violation of the system are calculated before and after installation of new GTs. The results are shown in Table 14. As can be seen in this table the active power loss and voltage violation are efficiently declined when new GTs are used. For example, the active power loss after using new GTs is 0.6587 (p.u.) for case 3 (max load). If these new GTs are installed in the system, this value will decrease about %2.17. Also, the value of voltage violation decreases about %2.26.
Table (12): Optimal multiobjective design using CSA for 32bus power system
Subsystem 1 
Gen 
G1 
G3 
G4 
G6 
GT1 
BSU 
 
 
 
 
√ 

NBSU 
√ 
√ 
√ 
√ 
 

30 
30 
20 
15 
15 

Location of BSU 
 
 
 
 
Bus 14 

Subsystem 2 
Gen 
G2  √ 30  
G5  √ 10  
GT2 

BSU 
√ 

NBSU 
 

15 

Location of BSU 
Bus 27 

Value of UEC (p.u . min) 
963.30 

Active power loss (p.u) 
0.6258 

Voltage violation (p.u) 
0.9316 
Table (13): The optimal cranking power paths of NBSUs
Location of BSU 
NBS unit 
optimal cranking paths 
Bus 14 
G1 
Bus: 14→11→9 
Bus 14 
G3 
Bus: 14→11→13 
Bus 14 
G4 
Bus: 14→11→3→1→4 
Bus 14 
G6 
Bus: 14→11→9→7 
Bus 27 
G2 
Bus: 27→15→28→29 
Bus 27 
G5 
Bus: 27→24→31 
Table (14): Obtained data of objective functions during power flow analysis
Objective function 
Active power loss 
Voltage violation 

Case 1 (min load) 
Without GTs 
0.6298 
0.9463 
With new GTs 
0.6030 
0.9241 

Case 2 (mid load) 
Without GTs 
0.6537 
0.9666 
With new GTs 
0.6250 
0.9365 

Case 3 (max load) 
Without GTs 
0.6733 
0.9733 
With new GTs 
0.6587 
0.9513 
6. Conclusions
In this paper, a new multiobjective design was proposed to find the best locations of GTs as the redundant blackstart units. During the optimization process, the optimal generators startup sequences with the optimal transmission paths are also determined. Unavailable energy capability, total active power loss and voltage deviation are considered as the objective functions and are minimized by using the CSA algorithm. The Pareto optimization method was applied to optimize the multiobjective problem and a group of answers, called Pareto optimal solution was obtained. The simulations results showed that active power loss and voltage violation are efficiently declined when new GTs are added to the system. Also, the unavailable energy capability decreased when these GTs were applied as redundant BSUs. It was also found that the CSA given in this paper has a high speed in convergence, especially when compared with other metaheuristic algorithms.
Appendix
The flowchart of whole mechanism of fuzzy based clustering is depicted in Fig. A, where Nmax, S, , Ni and Nj show the maximum size of repository, current size of repository, the distance between two neighbor clusters (i, j), the number of solutions in clusters Ciand Cj, respectively [35].
Fig (A): The flowchart of whole repository’s size controlling using fuzzy based clustering [35]
Nomenclature
output active power of the ith generator 

demanded active power at dth PQ bus 

the actual value of the voltage for ith PQ bus 

the actual value of the voltage for jth PQ bus 

the difference value of phase angle between bus i and bus j 

Number of generators 

Number of PQ buses 

output reactive power of the ith generator 

demanded reactive power at dth PQ bus 

the transfer susceptance between bus i and bus j 

the transfer conductance between bus i and bus j 

the value of the reference voltage 

the lower voltage limit of ith PQ bus 

the upper voltage limit of ith PQ bus 

maximum output power for NBS unit j 

required startup power for NBS unit j 

M 
Total number of NBS units 
certain time interval that NBS unit j need to be ready for receiving crank power 

maximum time interval for starting the NBS unit j 

time interval needed to energize the optimal path to NBS unit j 

generation capability of BS unit i for cranking the power through the optimal transmission path 

Starting time of NBS unit j 
[1] Submission date: 15, 05, 2018
Acceptance date: 17, 11, 2018
Corresponding author: Amin Khodabakhshian, Department of electrical engineering, Faculty of electrical engineering, University of Isfahan  Iran
[1] S.A. NezamSarmadi, S. Nourizadeh, S. Azizi, A. M. Ranjbar,” A power system buildup restoration method based on wide area measurement systems,” European Transactions on electrical Power, Vol. 21, pp. 712720, 2011.
[2] W. Ye, F. Xinyan, Y. Zheng, “Optimal power grid blackstart using fuzzy logic and expert system,” European Transactions on electrical Power, Vol. 19, pp. 969977, 2009.
[3] A. M. ElZonkoly, “Renewable energy sources for complete optimal power system blackstart restoration,” IET Gener Transm Distrib, Vol. 9, No. 6, pp. 531–539, 2015.
[4] D.H. Kim, Y.T. Yoon, S.S. Lee, S.K. Lee, “Power system restoration plan using the characteristics of scalefree networks,” European Transactions on electrical Power, Vol. 18, pp. 809819, 2008.
[5] W. Sun, C.C. Liu, L. Zhang, “Optimal Generator StartUp Strategy for Bulk Power System Restoration,” IEEE Trans Power Syst, Vol. 26, pp. 13571366, 2011.
[6] L.H. Fink, K.L. Liou, C.C. Liu, “From generic restoration actions to specific restoration strategies,” IEEE Trans Power Syst, Vol. 10, pp. 745751, 1995.
[7] S. Zeng, Z. Lin, F. Wen, G. Ledwich, “A new approach for power system blackstart decisionmaking with vague set theory,” Int J Electr Power Energy Syst, Vol. 34, pp. 114120, 2012.
[8] W.J. Liu, Z. Lin, F. Wen, G. Ledwich, “Intuitionistic fuzzy Choquet integral operatorbased approach for blackstart decisionmaking,” IET Gener Transm Distrib, Vol. 6, No. 5, pp. 378386, 2012.
[9] E. Lu, N. Wang, Z. Qin, H. Liu, Y. Hou, “Blackstart strategy for power grids including fast cut thermal power units,” Proc. IEEE Power and Energy Society General Meeting, pp. 15, 2013.
[10] C. Liu, M. Wu, Y. Deng, “Startup sequence of generators in power system restoration avoiding the backtracking algorithm,” Proc. IEEE Power and Energy Society General Meeting, pp. 15, 2013.
[11] C. Zhang, Z. Lin, F. Wen, G. Ledwich, Y. Xue, “Twostage power network reconfiguration strategy considering node importance and restored generation capacity,” IET Gener Transm Distrib, Vol. 8, pp. 91103, 2014.
[12] A. ElZonkoly, H Desouki, A. Farghaly, “Firefly based approach for optimal power system blackstart restoration,” Proc. EPE, pp. 1116, 2014.
[13] T. YU, J. Tong, K. W. Chan, “Study on microturbine as a backup power supply for power grid blackstart,” IEEE Power and Energy Society General Meeting, 2009.
[14] How hydroelectric energy works. [Online], http://www.ucsusa.org/clean_energy/ourenergy choices/renewableenergy/howhydroelectricenergy.html
[15] D. Grace, T. Christiansen, “Riskbased assessment of unplanned outage events and costs for combinedcycle plants,” Journal of Engineering for gas turbines and power, Vol. 135, No. 2, pp. 17, 2013.
[16] L. C. Ribeiro, L. Oliveira, E. L. Bonaldi, L. Silva, C. P. Salomon, J. G. Silva, G. L. Torres, “Automatic system for failure detection in hydropower generators,” Journal of Power and Energy Engineering, Vol. 2, pp. 3646, 2014.
[17] Blackstart service business rules. [Online], http://www.ieso.ca/ Documents/ consult/ se16/ se16_DACPSWG20060216pjmblackstart.pdf
[18] W. Sun, C. Liu, S. Liu, “Black start capability assessment in power system restoration,” IEEE Power and Energy Society General Meeting, pp. 17, 2011.
[19] Y. Chou, C. Liu, Y. Wang, C. Wu, C. Lin, “Development of a Black Start Decision Supporting System for Isolated Power Systems, “IEEE Trans. Power Syst, Vol. 28, No. 3, pp. 22022210, 2013.
[20] N. Jenkins, J.B. Ekanayake, G. Strbac, “Distributed generation, “ The Institution of Engineering and Technology, London, United Kingdom, 2010.
[21] A. Cano, F. Jurado, “optimal location of biomassfuelled gas turbines in an electric system, “ IEEE power engineering society meeting, pp. 16, 2006.
[22] C. C. Liu, S. J. Lee, S. S. Venkata, “An expert system operational aid for restoration and loss reduction of distribution systems,” IEEE Trans. Power Syst., Vol. 3, No. 2, pp. 619626, 1988.
[23] A. S. Bretas, A. G. Phadke, “Artificial neural networks in power system restoration,” IEEE Trans. Power Delivery, Vol. 18, No. 2, pp. 11811186, 2003.
[24] C. Wang, V. Vittal, S. Kolluri, S. Mandal, “PTDFbased Automatic Restoration Path Selection,” IEEE Trans. Power Systems, Vol. 25, No. 3, pp. 16861695, 2010.
[25] W. Sun, C. C. Liu, “Optimal Transmission Path Search in Power System Restoration,” IREP SymposiumBulk Power System Dynamics and Control –IX (IREP), pp. 15, 2013.
[26] Y. Fukuyama, H. Endo, Y. Nakanishi, “A hybrid system for service restoration using expert system and genetic algorithm,” Intell. Syst. Appl. Power Syst., pp. 394–398, 1996.
[27] D. Wang, L. Song, Y. Zhang, Z. Lu, H. Li, “The Bilevel programming optimization method of extended blackstart schemes considering safety factors during restoration process,” China International Conference on Electricity Distribution, pp. 15, 2016.
[28] T. D. Sudhakar, N. S. Vadivoo, S. M. R. Slochanal, S. Ravichandran, “Supply restoration in distribution networks using Dijkstra’s algorithm,” International Conf. on Power System Technology (POWERCON), pp. 640645, 2004, 640645.
[29] R. Askarzadeh, “A novel metaheuristic method for solving constrained engineering optimization problems: Crow search algorithm,” Computers and Structures, Vol. 169, pp. 112, 2016.
[30] E. W. Dijkstra, “A note on two problems in connection with graph,” Numerische mathematic, Vol. 1, pp. 269271, 1959.
[31] R. H. Liang, J. C. Wang, Y. T. Chen, W. T. Tseng, “An enhanced firefly algorithm to multiobjective optimal active/reactive power dispatch with uncertainties consideration,” Int J Electr Power Energy Syst. Vol. 64, pp. 10881097, 2014.
[32] T. Niknam, M. Bornapour, A. Ostadi, A. Gheisari, “Optimal planning of Molten Carbonate Fuel Cell Power Plants at distribution networks considering Combined Heat, Power and Hydrogen production,” Journal of Power Sources, Vol. 239, pp. 513526, 2013.
[33] M. Bornapour, R. Hooshmand, “An efficient scenariobased stochastic programming for optimal planning of combined heat, power, and hydrogen production of molten carbonate fuel cell power plants,” Energy, Vol. 83, pp. 734748, 2015.
[34] A. Khodabakhshian, M. R Esmaili, M. Bornapour,” Optimal coordinated design of UPFC and PSS for improving power system performance by using multiobjective water cycle algorithm,” Int J Electr Power Energy Syst, Vol. 83, pp. 124133, 2016.
[35] A. ShabanpourHaghighi, A. R. Seifi, T. Niknam, “A modified teaching–learning based optimization for multiobjective optimal power flow problem,” Energy Conversion and Management, Vol. 77, pp. 597607, 2014.
[36] S. Agrawal, B. Panigrahi, M. K. Tiwari, “Multiobjective particle swarm algorithm with fuzzy clustering for electrical power dispatch,” Evol Comput IEEE Trans, Vol. 12, pp. 529541, 2008.
[37] X. S. Yang, “Metaheuristic optimization,” Scholar pedia, Vol. 6, 2011.
[38] IEEE 39Bus power system test cases, [Online], Available: http:// electrica.uc3m.es/ pablole / varios_e.htm.
[39] Wang, V. Vijay, K. Sun, “OBDDbased sectionalizing strategies for parallel power system restoration,” IEEE Trans Power Syst,