You are on page 1of 14

Journal of Intelligent & Fuzzy Systems 16 (2005) 119131 IOS Press

119

A new approach for distribution state estimation based on ant colony algorithm with regard to distributed generation
T. Niknama,, A.M. Ranjbara and A.R. Shiranib
a b

Sharif University of Technology, Tehran, Iran Niroo Research Institute, Tehran, Iran

Abstract. Technology enhancement of Distributed Generations, as well as deregulation and privatization in power system industry, shows a new perspective for power systems and subsystems. When a substantial portion of generation is in the form of dispersed and small units, a new connection pattern emerges whereby the dispersed units are embedded in reticulation infrastructure. Now, the ow of power is no longer the same as in the conventional systems, since the dispersed generating plants contribute with generation also at the distribution grids level. Connection of generation to distribution grids cannot effectively be made, unless the some especial control and monitoring tools are available and utilizable. State estimation in these kinds of networks, often called mixtribution, is the preliminary and essential tool to fulll this requirement and also is the subject of this article. Actually, state estimation is an optimization problem including discrete and continuous variables, whose objective function is to minimize the difference between calculated and measured values of variables, i.e. voltage of nodes, and active/reactive powers in the branches. In this paper, a new approach based on Ant Colony Optimization (ACO) is proposed to solve this optimization problem. The feasibility of the proposed approach is demonstrated and compared with methods based on neural networks and genetic algorithms for two test systems. Keywords: Distributed generation, state estimation, Ant Colony Optimization (ACO)

1. Introduction Return back to 20 years ago, when vertical structure was the dominant structure for power system industry around the world, large-scale power plants were the best choice for system expansion due to their numerous advantages, such as low investment, operation and maintenance cost. As the capacities of these kinds of power plants were high, they were usually connected to transmission systems, and consequently to load centers, where distribution networks were responsible for supplying the loads. Now, the situation has been changed; the vertical structure is going to be eliminated all around
Corresponding author. Tel.: +98 21 8079395; Fax: +98 21 8590174; E-mails: tniknam@mehr.sharif.edu; taher nik@yahoo. com.

the world by means of unbundling of generation, transmission and distribution systems. Privatization and deregulation create this opportunity that the share of private sector is going to be increased in power system investment, and so give this possibility to the customers to be responsible for their own consumption with new types of generations, which have small size and user friendly performance. Utilization of these kinds of generations in power systems, again, changed the philosophy of classication of power system sub-sectors; now, the ow of power is no longer the same as in the conventional systems, since distribution systems have some kinds of generations by themselves. On the other hand, refer to huge investment in these kinds of generations, it is estimated that about 25% of total electricity generation will be produced by these kinds of generations till 2020 [14]. Therefore, development of new tools and algorithms to guarantee the

1064-1246/05/$17.00 2005 IOS Press and the authors. All rights reserved

120

T. Niknam et al. / A new approach for distribution state estimation based on ant colony algorithm

performance of the whole system will be an essential requirement, which must be taken into consideration rmly. In this paper, state estimation in a distribution system incorporating DGs has been reviewed. It is clear that while a huge number of small size generations are connected into distribution systems, appropriate monitoring and controlling algorithms are required to control the state variables of the system and minimize the operational cost of the whole network and at the same time the general security of the system should be taken into consideration. Distribution systems usually have a large number of nodes and connections, which need to be monitored and controlled. For this purpose, two methods can be applied. The rst one is based on installation of monitoring and controlling devices in each node while the second one is based on installation of these equipments in selected nodes. The rst method incorporates the measuring devices, which are equipped with some types of modem to send and receive the data. These measuring devices are going to be used in most of the new distribution systems to facilitate locally data logging and generations/consumptions management. The second method, due to lower number of monitoring and control devices, is much cheaper than the rst one, and it can be implemented in both new and old distribution systems; however, an estimation algorithm is needed to extend the results to the whole system. It is clear that, because of need to a mathematical estimation in this method, the accuracy is less than the rst one. This paper presents an algorithm for state estimation of distribution systems, where limited measuring devices are available. State estimators have been developed to provide the necessary data for real time control of transmission networks. They are typically based on the weighted least squares approach, which solves the system states by minimizing the mean-square-error of an overdetermined system of equations. A number of Distribution State Estimation (DSE) methods have been developed in distribution systems, which are divided into two main categories [1,2,5,7,8,11,12,19,20]: Statistical methods, which usually use an iterative convergence method. Load adjustment state estimation, which usually utilize sensitivity analysis. In the conventional methods belonging to both categories, it is assumed that the objective functions and constraints should be continuous and differentiable.

Existence of distributed generations, as well as SVCs and transformer tap changers with discrete performances, causes that these methods could not be easily used when a lot of discrete variables are incorporated. It has been revealed that, when there are some optimization problems where ordinary mathematical methods cannot be used easily, evolutionary methods and expert systems, such as neural networks and genetic algorithms would be good alternatives. Recently, a new evolutionary global optimization technique known as Ant Colony Optimization (ACO) has become a candidate for many optimization applications. The ant colony optimization has been used to solve several combinatorial optimization problems such as the traditional ones like Traveling Salesman Problem (TSP), Quadratic Assignment Problem (QAP), Job Shop Scheduling Problem (JSP), Single Machine Total Tardiness Problem (SMTTP), or some application of power system problems like unit commitment, economic dispatch, hydroelectric generation scheduling, reactive power pricing in deregulated system, voltage and var control in distribution systems [3, 4,6,9,10,13,1518]. In this paper, a new approach based on ant colony algorithm for a practical distribution state estimation including DGs, SVCs and Voltage Regulators (VRs) is presented. In this method, DGs and loads, which do not have constant outputs, are considered as state variables in which the difference between measured and calculated values is assumed as the objective function. Following this section, in Section 2, distribution state estimation problem is formulated. Unbalanced threephase power ow is described in Section 3. Modeling of distributed generation is proposed in Section 4. In Sections 5 and 6, ant colony system mechanism and the proposed ant colony algorithm are presented, respectively. Implementing the proposed approach to distribution state estimation is shown in Section 7. Finally, in Section 8, the feasibility of the proposed approach is demonstrated and compared with methods based on neural networks and genetic algorithms for two test systems.

2. Distribution state estimation with regard to distributed generation From a mathematical point of view, the state estimation problem is an optimization problem with equality and inequality constraints. The objective function is the summation of difference between the measured and

T. Niknam et al. / A new approach for distribution state estimation based on ant colony algorithm

121

calculated values. Distribution state estimation considering DGs can be expressed as follows:
n

The number of measurements in distribution systems is usually less than state variables. In order to have a unique answer, these assumptions should be made: Transmission lines and switches status are known. A contracted load and distributed generation values are known at each node. Voltage and current at the substation bus (main bus) are known. If outputs of DGs and loads are xed, outputs and power factors will be available. If outputs of DGs and loads are variable, the average outputs, standard deviations and power factors can be obtained. Set points of VRs and local capacitors are known.

Min f (X) =
i

= i (zi hi (X))2

X = [PG , PLoad ]
N 1 2 PG = [PG , PG , . . . , PG g ] 1 2 NL PL = [PLoad , PLoad , . . . , PLoad ]

s.t
i PG,min i PLoad,min i PG i PG,max i = 1, 2, 3, . . . , Ng

(1)

i PLoad

i PLoad,max i = 1, 2, 3, . . . , NL

ij ij PLine < PLine,max

T api min 0 Qi c

T api

T api i = 1, 2, 3, . . . , Nt max

3. Unbalanced three phase power ow In unbalanced three-phase power ow, the following components are modeled by their equivalent circuits in terms of inductance, capacitance, resistance and injected current. a) Distributed generations: DGs are modeled as PQ nodes. b) Transformers: transformers are modeled as equivalent circuit with ctitious current injections. c) Capacitors: capacitors are represented by their equivalent injected currents. d) Demands or loads: system loads are basically considered asymmetrical because of single-phase loads and unequal three phase loads. In this paper, a network-topology-based on threephase distribution power ow algorithm is used. Two matrices are used to obtain the power ow solution, which are the Bus Injection to Branch Current (BIBC) and the Branch Current to Bus Voltage (BCBV) matrices [18].

Qi i = 2, 2, 3, . . . , Nc c,max

where: X: state variables vector including loads and DGs outputs. zi : measured values. i : weighting factor of the i th measured variable. hi : state equation of the i th measured variable. m: number of measurements. Ng : number of DGs with variable outputs. NL : number of loads with variable outputs. Nc: number of capacitors installed along the feeder. Nt : number of VRs installed along the feeder. Qi , Qi c c,max : reactive power and maximum reactive power of the i th capacitor, respectively. i i i PG , PG,max and PG,min : active power, maximum and minimum active power of the i th DG, respectively. i i i PLoad , PLoad,max and PLoad,min: active power, maximum and minimum active power of the i th load, respectively. ij ij PLine and PLine,max : transmission line ow and maximum transmission line, respectively. T api : tap position of VRs T api and T api : minimum and maximum tap, max min respectively. In this paper, it is assumed that capacitors and VRs, which change stepwise and are installed along feeders, are controlled locally. During the search procedure, change of state variables (loads and DGs outputs) may cause change of tap positions and capacitor banks, which in consequence cause that the objective function changes non-continuously.

4. Distributed generation modeling Depending on the contract and control status of a generator, it may be operated in one of the following modes: In parallel operation with the feeder, i.e., the generator is located near the loads and designed to supply a large load with a xed real and reactive power output.

122

T. Niknam et al. / A new approach for distribution state estimation based on ant colony algorithm

Constant output power at a specied power factor. Constant output power at a specied terminal voltage. The generation nodes in the rst two cases can be well represented as PQ nodes. The generation nodes in the third case must be modeled as PV nodes. T. Niknam and A.M Ranjbar have presented an approach based on compensation method to model the generator as a PV node [18]. In this paper, DGs are considered as PQ nodes.

ij (k + 1) = ij (k) + ij

(3)

Where: ij : intensity of pheromone between nodes i and j, Lij : length of path between nodes i and j, : a coefcient such that (1- ) represents the evaporation of trail between time k and k + 1. 1 and 2 : control parameters for determining weights of trail intensity and length of path.

6. The proposed ant colony algorithm 5. Ant colony system mechanism Ants are insects, which live together. Since they are blind animals, they nd the shortest path from nest to food with the aid of pheromone. The pheromone is the chemical material deposited by the ants, which serves as a critical communication media among ants, thereby guiding the determination of the next movement. On the other hand, ants nd the shortest path based on the intensity of pheromone deposited on different paths. For better understanding,assume that ants want to move from A to B and vice versa, to obtain food (Fig. 1). At rst, if there is no obstacle, all of them will move along the straight path (Fig. 1a). At the next stage, assume that there is an obstacle; in this case, ants will not be able to follow the original trial in their movement. Therefore, randomly, they turn to the left (ACB) and to the right (ADB) (Fig. 1b). Since ADB path is shorter than ACB path, the intensity of pheromone deposited on ADB path is more than the other path. So ants will be increasingly guided to move on the shorter path (Fig. 1c). This behavior forms the fundamental paradigm of ant colony system. As it was indicated in Fig. 1, the intensity of deposited pheromone is one of the most important factors for ants to nd the shortest path. Therefore, this factor should be used to simulate the behavior of ants. Generally, the following factors are used to simulate ant systems: Intensity of pheromone Length of path To select the next path, the state transition probability is dened as follows: (ij )1 (1/Lij )2 (2) Pij = (ij )1 (1/Lij )2 After selecting the next path, trail intensity of pheromone is updated as: Up to now, a number of studies based on ant colony algorithm have been carried out in order to solve some optimization problems such as TSP, unit commitment and etc; however, the search domains in these studies are in discrete form. In addition to that, a number of methods based on ant colony have been presented to solve the problem in both continuous and discrete search domains. The main problem associated with these methods is that the time of convergence increases drastically when the number of variables is increased. In this section, a new approach based on ant colony algorithm will be presented to solve optimization problems in continuous and discrete domains simultaneously. In order to follow this goal, a number of N colonies, each of which contains M individuals and a Master are considered. The ant, which has the best position (minimum value of objective function) in each colony, is known as the Master of the colony. Now, to choose a movement direction, each of colonies needs to nd the best local and global positions as follows: 6.1. Finding the best local position Suppose the ith colony wants to change its position. At rst, the transition probabilities between the Master and the rest of the ants of colony are calculated as indicated in Eq. (4): [PLn ]i = [PLi1 , PLi2 , . . . , PLiM ]1M PLij = (Lij )1 (1/Lij )2
M j=1

(4)

(Lij )1 (1/Lij )2

where: Pij is the transition probability between the i th Master and the jth individual in the i th colony. Cumulative probabilities are calculated as:

T. Niknam et al. / A new approach for distribution state estimation based on ant colony algorithm

123

Fig. 1. An example of nding the shortest path by ants.

[CLn ]i = [CL1 , CL2 , . . . , CLM ]1M where CL1 = PLi1 CL2 = CL1 + PLi2 CL3 = CL2 + PLi3 ... ... CLj = CLj1 + PLij CLM = CLM1 + PLiM In the above equation, C Lj is the cumulative probability for the jth individual. The roulette wheel is used for stochastic selection of the best local position as follows: A number between 0 and 1 is randomly generated and compared to the calculated cumulative probabilities. The rst term of cumulative probabilities (C Lj ), which is bigger than the generated number, is selected and the associated position is considered as the best local position. 6.2. Finding the best global position Finding the best global position is similar to nding the best local position. The main difference between them is that in this case, Masters are considered to calculate transition probabilities. In other words, to choose the best global position, it is necessary to calculate the transition probability between each pair of Masters. 6.3. Determination of the next position The movement direction for each ant is obtained from a linear combination of the best global and local (5)

positions. All ants at each colony should be moved together along the specied direction. Figure 2 shows the above-mentioned process graphically.

7. Implementing the proposed ACO to distribution state estimation This section presents the implementation of the proposed algorithm to solve distribution state estimation problem. It should be noted that the state variables are loads and DGs values whose outputs are variable, rather than voltage or current as used by conventional state estimations. Step 1: Generating the initial population and trail intensity for Masters An initial population of Masters of colonies, X i , which must meet constraints, is generated randomly. At initialization phase, it is assumed that trail intensities between each pair of Masters are the same. Master Population = [X 1 , X 2 , . . . , X N ] Xi = [PG , PLoad ]1n
N 1 2 PG = [PG , PG , , PG g ] 1 2 NL PL = [PLoad , PLoad , . . . , PLoad ]

(6)

Xi min (j) Xi (j) Xi max (j) Xi (j) = rand() (Xi max (j) Xi min (j)) Xi min (j) j = 1, 2, 3, ..., n Master Trail Intensity = [Gij ]N N where N is the number of the colonies and n is the number of variables. rand() is a uniform random generator function.

124

T. Niknam et al. / A new approach for distribution state estimation based on ant colony algorithm

j * *
* * * *

Global
* * * * * * *

Local
*

Movement
* * * 1 2 *
3

* * * * *

* *

M *

N: Number of colonies M: Number of ants in each colony : Master ant

*: Ant in colony Global: The best global position Local: The best local position Movement: Movement direction of colony
Fig. 2. Determination of the movement direction for ant colony.

Step 2: Generating the initial population and trail intensity for ants in each colony In this step, an initial population is generated randomly for each ant colony. Also local trail intensities between each pair of ants in each colony are generated. Local Population = [Y 1 , Y2 , . . . , YM ] Xi (j) Yi (j) Xi (j) + Yi = [PG , PLoad ]1n
N 1 2 PG = [PG , PG , . . . , PG g ] 1 2 NL PL = [PLoad , PLoad , . . . , PLoad ]

(7)

Yi (j) = rand() (Yi max (j) Yi min (j)) Yi min (j) j = 1, 2, 3, . . . , n Local Trial Intensity = [Lij ]MM In this equation, M is the number of the ants in each colony, is the radius of local search area and n is the number of variables. rand() is a uniform random generator function. After generating the local population, the objective function is calculated for these individuals and the cor-

responding Master. The position, which has the minimum objective function, is selected as the Master of the colony. Step 3: Determination of the next position Assume that the ith colony wants to determine its next position. As mentioned before, the movement direction of each ant is a linear combination of the best global and local positions, which can be selected as: A). Selection of the best global position As stated before, the Masters are representatives of the colonies; therefore, the best global position is found based on them. Since L ij is not known in state estimation optimization problem, we can dene its inverse as follows: 1/Lij = Gij = f (X i ) f (X j ); j = i (8)

f (X i ) and f (X j ) are the objective function values of the state estimation problem for the i th and the jth Masters. Transition probabilities between the i th and the rest of the Masters are dened as:

T. Niknam et al. / A new approach for distribution state estimation based on ant colony algorithm

125

Read data include transmission line status, DGs and loads status, average and standard deviation of loads and DGs

Read real and pseudo measured values

Generate an initial population and trail intensities for Masters based on DGs and load values

Generate an initial population and trail intensities for each ant of one colony based on DGs and Load values

Calculate local transition probabilities based on trail intensities and different costs between each pair of ants in one colony

Calculate global transition probabilities based on trail intensity and different costs between each pair of Masters

Select the local position based on roulette wheel

Select the global position based on roulette wheel

Update trail intensities

Update trail intensities

Calculate and determine the next position based on the best local and global positions

Calculate values of measured points based on DGs and Loads values No Is convergence condition satisfied? Yes Stop and print results.
Fig. 3. Flowchart of the proposed algorithm.

PGij =

(Gij )1 (Gij )2

N (Gij )1 (Gij )2
j=1

j=i

j = 1, 2, . . . , N ; i = j

(9)

The cumulative probabilities for Masters are calculated based on calculated transition probabilities. The

best global position is selected by roulette wheel. After selecting the global position, global trail intensities are updated as follows: Gij = PGij (10) Gij (k + 1) = Gij (k) + Gij B). Finding the best local position Finding the best local position is similar to nding the best global one and is as follows:

126

T. Niknam et al. / A new approach for distribution state estimation based on ant colony algorithm

Fig. 4. Single Line Diagram of IEEE 34 bus.


Objective function value ACO

0.2 0.15 0.1 0.05 0

Table 1 Characteristic of generators Average of output (kW) Standard deviation (%) Power factor G1 65 0 0.8 G2 75 10 0.8 G3 95 10 0.8

Iteration Objective function value 0.0300

After selecting the best local position, local trail intensities are updated as follows: Lij = PLij Lij (k + 1) = Lij (k) + Lij (12)

11

13

15

17

GA

19

0.0200 0.0100 0.0000 1 21 41 61 81 101 Iteration

C). Determination of the next position After selecting the best local and global position, the next position is determined as follows: X i (k + 1)=X i (k)+rand() (Y Local X i (k)) +rand() (X Global X i (k)) (13) In above equation, Y Local and X Global are the best local and global positions, respectively. rand() is a uniform random generator function. It must be noted that, in the new position, the constraint restriction must be completely satised. Step 4: Check of convergence After all of the ant colonies found their next positions, the convergence is checked by:
N

Fig. 5. Convergence characteristic of GA and ACO for the best solution.

At rst, transition probabilities between the i th Master and the local ants are calculated as: Lij = f (X i ) f (Y j ); PLij = (Lij )1 (Lij )2
M j=1

(11)

(Lij )1 (Lij )2

Xi
i=1

k+1

Xi

k 2

<

(14)

j = 1, 2, . . . , M ; f (X i ) and f (Y j ) are the objective function values of the state estimation problem for the i th Master and the jth ant in the ith colony. The cumulative probabilities are calculated and the best local position is selected by roulette wheel.

If the convergence condition is satised, the task is complete and if not, the process must be repeated from step 3. Figure 3 shows the complete owchart of the process. The variations range of M and N are between 4 to 10 and 10 to 100, respectively and depend on the number of unknown variables (dimension of vectorX).

T. Niknam et al. / A new approach for distribution state estimation based on ant colony algorithm Table 2 Characteristic of variable loads Average of output (kW) Standard deviation (%) Power factor Location
Objective function value

127

Table 3 Comparison of measured and estimated values L2 90 10 0.75 34 Method Actual value 70 70 70 80 80 80 90 90 90 75 75 75 Calculated value Best Worst solution solution 69.95 70.2 69.4 69.4 69.78 72.234 80.02 79.5 80.3 80.3 79.98 78.64 89.84 89.2 89.2 89.2 90.5 92.378 75.213 74.3 74 74 76.4 77.63

L1 80 10 0.75 25

G2

6.00E-07 5.00E-07 4.00E-07 3.00E-07 2.00E-07 1.00E-07 0.00E+00

Min=5.87*10e(-8) Max=48*10e(-8)

ACO

G3

L1

L2

ACO NN GA ACO NN GA ACO NN GA ACO NN GA

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Execution Objective function value

Table 4 Comparison of the execution time and objective function values Method Objective function 6.84* 10(5) 6.84* 10(5) 1.18* 10(5) 887* 10(5) 5.87* 10(8) 48*10(8) Execution time(s) 0 0 40 120 4 14

0.0100 0.0080 0.0060 0.0040 0.0020 0.0000

Min=1.18*10e(-5) Max=877*10e(-5)

GA

NN GA ACO

Best solution Worst solution Best solution Worst solution Best solution Worst solution

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
Execution

Table 5 Comparison of average and standard deviation for different executions Method ACO GA NN Average 32*10(8) 294* 10(5) 6.84* 10(5) Standard deviation 14*10(8) 311* 10(5) 0

Fig. 6. Comparison of objective function values for different executions.

8. Simulation In this section, the proposed method is applied to the distribution state estimation problem on two distribution test systems. It is assumed that the following information is available. Value of output for each of the constant loads and DGs. Average value and standard deviation for each of the variable DGs and loads. Values of the measured points Power factors of Loads and DGs Set points of VRs and local capacitors In following, results for two cases are presented. Case 1: IEEE 34 bus radial test feeders Figure 4 shows the IEEE 34 bus radial distribution test feeders whose associated specications are presented in [22]. For this system, it is assumed that there are three DGs (two variable and one constant) connected at buses 9, 23

and 27, respectively whose specications are presented in Table 1. There are also two variable loads whose specications are demonstrated in Table 2. It is assumed that there are three measurement devices installed on buses 1, 15 and 25. These devices are Ammeters and Wattmeters. Tables 3, 4 and 5 and Figs 5 and 6 show the comparison between the results of the proposed method, Genetic Algorithm and Neural Networks, while their simulation parameters are presented in appendix. Case 2. A realistic 23 bus 20 Kv network The proposed method is also applied to a rural network as shown in Fig. 7. This system is used to supply power demand in the village located in the north of Iran. Line and load characteristics are shown in Tables 6 and 7, respectively. Line Impedance Matrix is presented in Eq. (15). As there is no DG in this network currently, two typical DGs have been considered on buses 13 and 21 whose specications have been presented

128

T. Niknam et al. / A new approach for distribution state estimation based on ant colony algorithm

Fig. 7. Single Line Diagram of the rural network. Table 6 Line characteristics No 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 From 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 10 17 17 19 18 21 22 To 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 Length (m) 40 280 140 120 330 725 210 210 55 60 1000 1020 870 865 865 1400 1700 70 70 1060 1500 520

Hidden Layer Input Layer Output Layer

Fig. A1. The structure of neural network.

9. Discussion As shown in Tables 3, 4, 5, 6 and 7 and Figs 5 and 6, the proposed method can be used to estimate state variables in distribution networks. The results, presented in these Tables and Figures, can be summarized as follows: The execution time of the proposed method is signicantly short with respect to GA and gives a general idea that the method can be implemented without any restriction in practical networks. The execution time of Neural Network is less than the proposed method, however, it is more feasible than neural networks, which needs a lot of training patterns, as the proposed algorithm requires only network data and measured values. Presented method is very precise. In other words, not only does this method reach to better optimal solution with respect to other methods, but also the standard deviation for different trials (with regard to GA) is very small (Fig. 6 and Table 5).

in Table 8. In this system, there are 2 variable loads whose characteristics are shown in Table 9. ZLine (/m) = (1e 4) 7 + j7 0.2 + j.15 0.2 + j.15 0.2 + j.15 7 + j7 0.2 + j.15 (15) 0.2 + j.15 0.2 + j.15 7 + j7 It is assumed that there are three measuring devices installed on buses 1, 11 and 19. These devices are Ammeters and Wattmeters. A comparison between the proposed algorithm (ACO), Genetic Algorithm and Neural Network is shown in Tables 10 and 11.

T. Niknam et al. / A new approach for distribution state estimation based on ant colony algorithm Table 7 Load characteristics No 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 Pa(Kw) 0.00 105.00 83.33 83.33 83.33 83.33 83.33 83.33 83.33 83.33 105.00 105.00 83.33 83.33 21.00 333.33 133.33 83.33 105.00 105.00 50.00 0.00 105.00 Qa(Kvar) 0.00 78.75 62.50 62.50 62.50 62.50 62.50 62.50 62.50 62.50 78.75 78.75 62.50 62.50 15.75 250.00 100.00 62.50 78.75 78.75 37.50 0.00 78.75 Pb(Kw) 0.00 114.45 90.83 90.83 90.83 90.83 90.83 90.83 90.83 90.83 114.45 114.45 90.83 90.83 22.89 363.33 145.33 90.83 114.45 114.45 54.50 0.00 114.45 Qb(Kvar) 0.00 85.84 68.13 68.13 68.13 68.13 68.13 68.13 68.13 68.13 85.84 85.84 68.13 68.13 17.17 272.50 109.00 68.13 85.84 85.84 40.88 0.00 85.84 Pc(Kw) 0.00 95.55 75.83 75.83 75.83 75.83 75.83 75.83 75.83 75.83 95.55 95.55 75.83 75.83 19.11 303.33 121.33 75.83 95.55 95.55 45.50 0.00 95.55 Qc(Kvar) 0.00 71.66 56.88 56.88 56.88 56.88 56.88 56.88 56.88 56.88 71.66 71.66 56.88 56.88 14.33 227.50 91.00 56.88 71.66 71.66 34.13 0.00 71.66

129

The proposed method converges fast (Fig. 5). The simulation results show that estimation errors are in acceptable levels (Tables 4 and 11). The method can estimate the appropriate target system condition even with measuring devices errors (Tables 2 and 5). The proposed algorithm can estimate appropriate loads and DGs output values at each node with limited measurement points in distribution networks. The method can be applied to a wide variety of similar optimization problems with non-differential and non-continuous objective functions and constraints.

Table 8 Characteristic of generators Average of output (kW) Standard deviation (%) Power factor G1 850 10 0.8 G2 950 10 0.8

Table 9 Characteristic of variable loads Average of output (kW) Standard deviation (%) Power factor Location L1 300 10 0.75 16 L2 250 10 0.75 19

10. Conclusion As the number of DGs grows, their impacts on power system have to be studied more. One of the most important issues in distribution systems is distribution management system (DMS), which can be affected by DGs. State estimation in DMS plays a key role in estimating the system real-time state. An efcient approach to estimate distribution state variables in the presence of DGs is presented in this paper. The results of simulations indicate that the method can estimate target system conditions accurately. Also, the proposed method could be applied to a wide variety of similar problems.

The execution time of proposed method is remarkably short and gives a general idea that the method can be implemented without any restriction in practical networks. The following subjects can be considered in future works: Estimation of DGs outputs when they are modeled and controlled as PV nodes. Estimation of tap position of VRs. Estimation of DGs outputs when they are controlled and modeled separately. In this situation, voltage magnitude and reactive power for each phase of DGs can be controlled separately.

130

T. Niknam et al. / A new approach for distribution state estimation based on ant colony algorithm Table 10 Comparison of measured and estimated values Method ACO NN GA ACO NN GA ACO NN GA ACO NN GA Actual value 800 800 800 920 920 920 310 310 310 240 240 240 Calculated value Best solution Worst solution 799.9 801.2 798.1 798.1 796.1 810 920.3 918.4 921.123 921.123 926.42 928.78 309.71 311.58 312.4 312.4 311.56 317.75 239.56 238.89 246.2 246.2 235.1 247.63

A3. Neural Network In this paper, a MLP neural network with BP learning technique is used whose structure is shown in Fig. A1. Simulation conditions are: Number of total patterns = 2000; Number of training patterns = 1500; Test data = 500; Number of hidden layers = 1; Number of neurons in hidden layers = 4; Number of neurons in input layers = Number of measurement devices Number of neurons in output layers = Number of DGs + Number of Loads

G1

G2

L1

L2

Table 11 Comparison of the execution time and objective function values Method NN GA ACO Best solution Worst solution Best solution Worst solution Best solution Worst solution Objective function 45.56*10(5) 45.56*10(5) 488.1*10(5) 986.45*10(5) 2.523*10(8) 284*10(8) Execution time(s) 0 0 35 70 3 12

Acknowledgments The authors gratefully acknowledge the helps from Dr. Mahmoud Fotohi, Mehran Mirjafari, Babak Mozafari and Sohrab Amini.

References
K. Ghosh, D.L. Lubkeman, M.J. Downey and R.H. Jones, Distribution circuit state estimation using a probabilistic approach, IEEE Trans. Power Syst. 12 (Feb., 1997), 4551. [2] P.S. Meliopoulos and F. Zhang, Multiphase power ow and state estimation for power distribution systems, IEEE Trans. Power Syst. 11 (May, 1996), 939946. [3] A. Bauer, B. Bullnheimer, R.F. Hartl and C. Strauss, An ant colony optimization approach for the single machine total tardiness problem, Proceedings of the 1999 Congress on Evolutionary Computation, CEC 99 2 (July, 1999), 1450. [4] D. Merkle, M. Middendorf and H. Schmech, Ant colony optimization for resource constrained project scheduling, IEEE Trans. On Evolutionary Computation 6(4) (Aug., 2002), 333 346. [5] Roytelman and S.M. Shahidehpour, State estimation for electric power distribution systems in quasi real-time conditions, IEEE Trans. Power Delivery 8 (Oct., 1993), 20092015. [6] K. Krishnoiger and S.H. Cheraghi, Ant Algorithms: Review and Future Application, Industrial Engineering Research Conference, Orlando, May, 2002, USA. [7] M.E. Baran and A.W. Kelley, State estimation for real-time monitoring of distribution systems, IEEE Trans. Power Syst. 9 (Aug., 1994), 16011609. [8] M.D. Nimrihter, R.M. Ciric and A. Slijepcevic, Measurements and pseudo-measurements in distribution networks, IEEE Instrumentation and Measurement Technology Con. Budapest, Hungary, May, 2001, pp. 17911796, [9] M. Dorigo, G.D. Caro and L.M. Gambardella, Ants algorithms for discrete optimization, Articial Life 5(3) (1999), 137172. [10] N.S. Sisworahardjo and A.A. El Keib, Unit Commitment Using the Ant Colony Search Algorithm, 2002 Large Engineering Systems Conference on Power Engineering, LESCOPE 02, June, 2002, pp. 26. [1]

Appendix This appendix describes processes of the used ACO, GA and NN as well as illustrates their simulation conditions introduced in Section 8. A1. Ant Colony Algorithm N = 15; M = 4; 2 = 2; 1 = 4; = 0.9; A2. Genetic Algorithm In this paper, Integer strings instead of binary coding are used to represent value of variables, and include these processes: Representation and initialization Fitness function Reproduction operation Crossover operation Mutation operation

Simulation conditions are: Initial population = 1000; Selected Population = 100; Mutation = 4 Percent; Cross Over Probability = 0.2 to 0.3;

T. Niknam et al. / A new approach for distribution state estimation based on ant colony algorithm [11] R.C. Leou and C.N. Lu, National Sun Yat-Sen University, Taiwan, Improving feeder voltage calculation results with telemetered data, IEEE Trans. Power Delivery 11 (Oct., 1996), 1914 1920. S. Naka, T. Genji, T. Yura and Y. Fukuyama, A hybrid particle swarm optimization for distribution state estimation, IEEE Trans. on Power system 18(1) (Feb., 2003), 6068. S.J. Huang, Enhancement of hydroelectric generation scheduling using ant colony system based optimization approaches, IEEE Trans. On Energy Conversion 16(3) (Sep., 2001), 296 301. T. Ackerman, G. Anderson and L. Soder, Distributed Generation: A Denition, Elsevier Science, Dec., 2000, pp. 195204. T. Niknam, A.M. Ranjbar and A.R. Shirani, Optimization with ant colony algorithm, BARGH Journal of Electrical Science and Technology 36 1119. T. Niknam, A.M. Ranjbar and A.R. Shirani, A New Approach Based on Ant-Genetic Algorithm for Volt/Var Control in Distribution Network with Considering Distributed Generation, IEEE DRPT2004 Hong Kong, April, 2004. T. Niknam, H. Arabian and M. Mirjafari, Reactive Power Pric-

131

[12]

[18]

[13]

[19]

[20]

[14] [15]

[21]

[16]

[22]

[17]

ing in Deregulated Environments Using Novel Search Methods, (Vol. 7), Proceedings of 2004 International Conference on Machine Learning and Cybernetics, Aug., 2004, pp. 4234 4239. T. Niknam and A.M. Ranjbar, Impact of Distributed Generation on Distribution Load Flow, International Power System Conference Tehran, Iran, Oct., 2002. Y. Deng, Y. He and B. Zhang, A branch-estimation-based state estimation method for radial distribution systems, IEEE Trans. On Power Delivery 17(4) (Oct., 2002), 10571062. Y. Deng, Y. He and B. Zhang, Branch-estimation-based state estimation for radial distribution systems, 2000 IEEE Power Engineering Society Winter Meeting 4 (Jan. 2000), 2351 2356. Y.H. Hou, Y.W. Wu, L.J. Li and X.Y. Xiong, Generalized Ant Colony Optimization for Economic Dispatch of Power Systems, (Vol. 1), Power Con 2002 Proceedings of International Conference on Power System Technology, Oct. 2002, pp. 225229. W.H Kersting, Radial distribution test feeders, IEEE Trans. On Power Systems 6(3) (Aug. 1991), 975985.

You might also like