Intelligent algorithm for optimal meter placement and bus voltage estimation in ring main distribution system

Abstract

The advancement in power distribution system poses a great challenge to power engineering researchers on how to best monitor and estimate the state of the distribution network. This paper is executed in two stage processes. The first stage is to identify the optimal location for installation of monitoring instrument with minimal investment cost. The second stage is to estimate the bus voltage magnitude, where real time measurement is conducted and measured through identified meter location which is more essential for decision making in distribution supervisory control and data acquisition system (DSCADA). The hybrid intelligent technique is applied to execute the above two algorithms. The algorithms are tested with institute of electrical and electronics engineers (IEEE) and Tamil Nadu electricity board (TNEB) benchmark systems. The simulated results proves that the swarm tuned artificial neural network (ANN) estimator is best suited for accurate estimation of voltage with different noise levels.

Keywords

artificial intelligence / power distribution control / state estimation

Introduction

The excellence in power generation through distributed energy sources is to give more consideration on power distributed management system. Micro monitoring of power distribution feeder helps to achieve optimal operation of the distribution system. Proper placement of meters in all buses and lines will help to achieve micro monitoring of the distributed system. Identifying best possible location to place meters is termed as optimal meter placement problem. Optimal placement of meters will reduce the initial investment cost when considering the system consisting of a large number of buses.

Technical decisions made in distributed systems are based on the monitoring parameters through distribution supervisory control and data acquisition system (DSCADA), where state estimation is the major tool to identify the measurement error before any action. Extensive research has already been conducted in power system state estimation meter placement, where there is a possibility of acceptable measured data with limited pseudo measurements. But in the case of power distribution system, limited numbers of measurements are available with a large number of pseudo measurements. It is an arduous task to get an accurate estimated value, if the state estimator is not designed properly. The intelligent meter placement and hybrid state estimator are designed in this paper to achieve optimal operation and control of the power distribution management system. The algorithm for meter placement is formulated using particle swarm optimization, and the algorithm for bus voltage estimation is formulated using Swarm tuned artificial neural network (ANN). The formulated algorithm is verified with the institute of electrical and electronics engineers (IEEE) and Tamil Nadu electricity board (TNEB) benchmark systems.

Meter placement and state estimation techniques

The approach of estimating the monitoring data through SCADA was first developed by Scheppe [1] and a power system research group at MIT in 1975. The initial approach on power system estimation was proposed with weighted least sequence technique [2]. This is further developed by different techniques like simulated annealing (SA), genetic algorithm (GA), ANN, sparse triangular factorization (STA), linear programming (LP), and integer programming (IP) by different authors. And a detailed review of power system state estimation meter placement was made [3]. Later, in the year 1996, research was carried out to implement the state estimation meter placement in distribution system. Rule based meter placement algorithm [4] was proposed first for distributed feeders. The rules developed for meter placement were based on different observation for real time measurement and forecasted load data as pseudo measurements. The proposed rule is implemented on distribution energy control (DEC) workstation environment and tested using four different sized feeders. Then the research on state estimation meter placement was extended to different algorithms like GA [5], reverse branch current [6], ANN [7], multiple load flow [8], dynamic program [9], reconfigurable distribution automation and control (RDAC) [10], variance moment GA [11] and particle swarm optimization (PSO) [12]. The comparative study of different state estimation meter placement algorithm was discussed in Ref. [13].

The latest algorithm was developed for distribution state estimation meter placement by Singh et al. [14] using Bivariate Chebyshev bound technique. This proposed technique was to improve the quality of voltage and angle, and this estimation across a network was based on the sequential improvement of a bivariate probability index governing errors in voltage and angle at each bus. The algorithm was tested with 95-bus UK generic distribution system (GDS) for which the result is feasible but not the optimal solution. The impacts of distributed generation (DG) on distribution system create new monitoring, control and management issues. Designing new state estimation and optimal meter placement including DG in the system becomes significant in the scenario. In the Monte Carlo- dynamic programming (DP) algorithm [15] development by Carlo, the presence of DG was considered for meter placement. The algorithm was tested for Italian distribution network.

In previous work [16], an algorithm was developed for distributed distribution network state estimation based on back propagation neural network and tested for IEEE 37 bus system. The estimated bus voltage and real power was calculated by the proposed weighted least square estimation (WLSE) and ANN method described in their paper. The performance of the proposed neural network was compared with conventional WLSE and concluded for 50% improvement in real power accuracy and 40% improvement in bus voltage. The time comparison of proposed method was 40% better than WLSE. The proposed method incorporates bad data detection and system observability. The limitations in their work were that real time measurements were taken randomly and location of meter was not considered. Gelagaev et al. [17] developed two methods for the state estimation in distribution grid. They were weighted least squares (WLSs) and extended Kalman filter. Both estimators try to find the most probable state based on the available measurements. The outcome of the Kalman filter is that it requires less number of iterations and less calculation time, while the disadvantage of it is that some foreknowledge about the state is needed. Singh et al. [18] created a statistical framework to assess the suitability of various state estimation methodologies for the purpose of distribution system state estimation. The existing algorithms adopted in the transmission system SE were reconfigured for the distribution system. The performance of three SE algorithms WLS, weighted least absolute value estimator (WLAV) and schweppe huber generalised mestimator (SHGM) was examined and discussed in standard 12-bus and 95-bus UK-GDS network models. Baran and McDermott [19] discussed the application of state estimation for real-time monitoring of distribution systems. It shows that with the availability of new measurements from AMI, estimation of operating conditions on a distribution system can be estimated quite accurately on a feeder basis. Singh et al. [20] developed an efficient approach for distribution system state estimation (DSSE) based on the Gaussian mixture model (GMM) of loads as pseudo-measurements was presented and theoretically justified. The advantage of the approach is that all the load pdfs irrespective of their distributions are represented by GMM approximation followed by the appropriate reduction.

PSO algorithm for optimal meter placement

Identifying node, finding location for placement of remote terminal unit (RTU) and flow meter are designed as an optimal meter placement problem. It is not possible to place RTU and line flow meters at every node and line respectively, because of high initial investment cost. To achieve best operating state of the system, distribution state estimation (DSE) function collected information of network elements and a set of redundant measurements. The performance of the DSE function is primarily influenced by real-time data redundancy. The term redundancy refers to a surplus of measurements gathered from the supervised network in relation to a minimum number necessary to estimate all the state variables. An adequate redundancy was assessed through a metering system designed in which not only the number, but also the type and location of measurements were considered, mainly to meet the following requirements. The observability of DSE was accomplished for the entire system network, the gross error detection, identification, suppression and possibilities to check the reliability of the system. In the event of network configuration changes or temporary malfunction of the data acquisition system, implying in redundancy deterioration, observability and reliability requirements were still met to check the robustness of the system and to minimize the investment cost for data acquisition.

The optimization problem formulated [21,22] to achieve minimal meter placement is

$$\begin{gathered} \min \underset{i=1}{\overset{n}{\mathrm{\Sigma }}}\left\{\mathrm{\Sigma }(C_{RTU}+C_m)\right\}*X_i, \\ subjecttoperformancerequirem\mathrm{ent} \\ \end{gathered}$$

where C_RTU is the cost of RTU, C_m is the cost of flow meter. X_i is binary decision variable, X_i is equal to one, if meter is at bus i, X_i is equal to zero for all other conditions.

The performance requirements refer to the quantity, type and location of meters in the electrical network in order to guarantee a desirable performance of the real time monitoring process.

The intelligent search algorithm, formulated by particle swarm optimization was used to identify the optimal metering position. The PSO [23-25] algorithm which is utilized for various power and distributed system application yields accurate result. The solution algorithm for meter placement scheme is given below:

1) Enter the input bus and line data.

2) Set the constraints such as switch position, location of breakers, location of sectionalizes.

3) Enter the PSO input parameters such as number of particles, type of model, maximum velocity divisor and accuracy.

4) Calculate cost of planning RTU/meters.

5) Initialize swarm and start iteration.

6) Calculate the velocity vector.

7) Update the position vector.

8) If constraints are satisfied, go to 9 else go to 6.

9) Calculate the optimal positioning.

10) If iteration is satisfied, go to 11 else go to 3.

11) Check the observability of the system.

12) If the system is observable, display the result else go to 6.

13) End.

Bus voltage estimation

The operation and control action of power distribution system depends on the estimated measured value through meters/RTU as described in Fig. 1.

The most familiar technique in practice is the weight least square estimation. Many authors [17-20] discussed distribution state estimation technique with different aspects, with the application to benchmark distribution systems. This work is to improve the accuracy of the estimation by using PSO and ANN named as swarm tuned ANN estimator. The RTU/meters were placed in respective identified location by PSO meter placement. The real time measurements like bus voltage, line currents, line real power flow, line reactive power flow and load current were measured through respective meters and communicated to RTU, which communicates through wireless communication to the DSCADA. The estimator as a part of DSCADA estimated the bus voltage magnitude before approval for any action.

Pseudo-measurements

One of the major differences in the distribution state estimation is the deficiency in getting real time measurement. Only limited RTU/meters are there to monitor the real time measured value to the respective bus/line. The estimator needs data for all the buses and lines. To satisfy the above needs, pseudo-measurements modelling is necessary to feed the local data for estimation. However this is quite easy in power system state estimation because more measurement and few pseudo measurements are considered. To model the pseudo measurements, different approaches were followed by different authors [26,27]. The modelling of pseudo measurements in this work was based on the correlation approach discussed in Ref. [27]. The electrical transient analyzer program (ETAP)<FootNote>

Electrical Transient Analyzer Program (ETAP), www.etap.com

</FootNote> was used to run the load flow with various load conditions from time-time to get the load bus voltage magnitude.

Swarm tuned ANN estimator

The bus voltage magnitude of power distribution system was effectively used to decide the operating state of a system. State estimation techniques were used to estimate the bus voltage. Weight least square technique [28] is the method regularly used in power system to check the state of the system. The accuracy of WLSE is not satisfactory because of the inherent capability of filtering the bad data and computation time. Das and Naka et al. [7,29] discussed the intelligent estimation technique which avoided the mathematical calculation and hence reduced the memory space. This work describes the hybrid technique i.e. combining ANN with PSO. The fine tuning of weight is the major contribution of ANN. This process was conducted by PSO. The basic architecture of the single layer ANN is illustrated in Fig. 2.

The bus voltage magnitude of power distribution system was effectively used to decide the operating state of a system. State estimation techniques were used to estimate the bus voltage. Weight least square, the load current I_L, real power flow P_ij , reactive power flow Q_ij and line current I_ij were taken as input for training ANN. The bus voltage magnitude would be the output of ANN.

The global optimization technique [30-33] developed was based on the concept of flock of birds and a school of fish in order to guide the swarm of particle towards the most promising regions of the search space. The particles movement would be based more on velocity. The performance of each particle was measured according to a predetermined fitness function related to the application. The representation of particle in this work was the weight vector of ANN, including biases. The dimension of the search space was therefore the total number of weights and biases.

The iterative approach of PSO could be described by the following steps:

1) Initialize a population size, positions, velocities of agents and the number of weights and biases.

2) The current best fitness achieved by particle P is set as P_best. The P_best with best value is stored.

3) Evaluate the desired optimization fitness function f_p for each particle as the mean square error (MSE) over the given data set.

4) Compare the evaluated fitness value f_p of each particle with its P_best value. If f_p<P_best then P_best = f_p and best{best}_{x_p}=x_p\mathrm{.}{x}_px_p is the current co-ordinates of particle P’s best fitness so far.

5) The objective function value is calculated for new position of each particle. If a better position is achieved by an agent, P_best value is replaced by the current value. As in step 4, P_best value is selected amongst P_best values. If the new P_best value is better than the previous P_best values, the P_best is replaced by current P_best value and this value is stored. If f_b<P_best then P_best=P, where P_best is the particle having the overall best fitness in the swarm.

6) Change the velocity and the location of the particle according to Eqs. (1) and (2).

7) Fly each particle P according to Eq. (2).

8) If the maximum numbers of a predetermined iteration is exceeded, then stop; otherwise go to step 3 until it converges. In this work, 25 populations of weights were evolved for 200 generations.

(1)v_i=w{v}_{i-\mathrm{1}}+\mathrm{a}\mathrm{c}\mathrm{c}·\mathrm{r}\mathrm{a}\mathrm{n}\mathrm{d}()·(\mathrm{b}\mathrm{e}\mathrm{s}{\mathrm{t}}_{x_{\mathrm{p}}}-x_{\mathrm{p}})+\mathrm{a}\mathrm{c}\mathrm{c}·\mathrm{r}\mathrm{a}\mathrm{n}\mathrm{d}()·(\mathrm{b}\mathrm{e}\mathrm{s}{\mathrm{t}}_{x_{\mathrm{g}\mathrm{b}\mathrm{e}\mathrm{s}\mathrm{t}}}-x_{\mathrm{p}}),

where, acc is the acceleration constant that controls how far the particles fly from one another and rand returns a random number between 0 and l, v_i is the current velocity , v_i-1 is the previous velocity, x_p is the present location of the particle, x_pp is the previous location of the particle and i is the particle index. In step 5 the coordinates \mathrm{b}\mathrm{e}\mathrm{s}{\mathrm{t}}_{x_{\mathrm{p}}} and \mathrm{b}\mathrm{e}\mathrm{s}{\mathrm{t}}_{x_{\mathrm{g}\mathrm{b}\mathrm{e}\mathrm{s}\mathrm{t}}} are used to pull the particles towards the global minimization.

(2)x_{\mathrm{p}}=x_{\mathrm{p}\mathrm{p}}+v_i.

Test results

This section is categorized into two parts. The first section will execute the PSO meter placement algorithm to find the best location, while the second section is to estimate the bus voltage using swarm tuned estimator. The proposed algorithms were tested on modified IEEE and TNEB distribution system where the source code is written in MATLAB 6.1 [34,35]. The meter placement and monitoring simulation was done in PSCAD 4.1<FootNote>

PSCAD/EMTDC Users’ Manual, Manitoba HVDC Research Center, 2003

</FootNote> software package, whereas the analysis of Pseudo-Measurements done through ETAP 5.5 version.

PSO meter location

The algorithm tested with minimum number of nodes was limited to 50 as presented in Ref. [36]. The previous work on PSO meter placement algorithm [36] suffered from the disadvantages of poor performance if the number of nodes exceeded 50 nodes, where the inputs were normalized to zero to one. This limitation was overcome in this work. The feasibility of modified proposed algorithm used for N number of nodes was tested on modified IEEE 202 bus and TNEB 140 bus system.

Figure 3 represents the location of RTU with respect to the distance taking base as input supply from substation or grid for IEEE 202 ring main system. The minimum number of RTU needed to make the system observable was 68 with minimum cost of 111 units. The calculation of cost is based on one unit for RTU and 0.2 units for flow meters. The location of flow meters with respect to RTU position is displayed in Fig. 4. This graph is plotted based on the number of flow meters needed corresponding to respective RTU. Figure 5 depicts the location of RTU with respect to the distance, taking base as input supply from substation or grid for TNEB 140 ring main system. The minimum number of RTU needed to make the system observable is 51 with minimum cost of 51 units. The location of flow meters with respect to RTU position is demonstrated in Fig. 6. This graph is plotted based on the number of flow meters needed corresponding to respective RTU.

The IEEE 202-ring main distribution system was worked out with different possible simulation considering all constraints. In all tests it was assumed that there were no previously installed meters. The different combination of topology scenarios were considered and simulated with test system. The best possible combination (minimum RTU, minimum flow meter with minimum cost) for location of RTU and flow meter is visualized in Figs. 4 and 5. From the simulated results it was clearly identified that PSO is best suited for finding the location of RTU and meter with minimum investment meter cost. The simulated result showed that there is an option for minimizing the number of RTU to be installed. The low redundant metering systems obtained indicated that the proposed method is capable of finding solutions of very low cost with best location, while satisfying the specified constraints when compared with [9,21,37]. The computation time for IEEE 202 bus system is 5-10 minutes which is comparatively low when compared with that in Ref. [21].

Swarm tuned bus voltage estimation

The proposed algorithm was tested with radial distribution feeders with minimum number of nodes limited to 50 presented in Ref. [38]. The limitation in the algorithm due to error accuracy by limited inputs for training ANN is eliminated in this paper by considering load current as added input for training data and considering maximum constraints. The feasibility of modified proposed algorithm used for N number of radial and ring nodes was tested on modified IEEE 202 bus and TNEB 140 bus system.

In any distribution system, available measurements such as real power line flow (P_ij ), reactive power line flow (Q_ij ), line current (I_ij ), load current (I_L ) and bus voltage (V_i ) are available at the substation. The multi-layer feed forward neural network architecture was used to execute the algorithm. The input patterns at the input nodes of the network comprised of the measurement triplets (P, Q, I_L , I_ij ) and the output patterns at the output nodes consisted of the system bus voltage magnitudes at all the buses. To train the ANN, it was necessary to generate a number of input-output patterns at different loading conditions. By varying the K_w and K_var loads in the system within a certain range with respect to the base loading condition, the training data for different loading condition was achieved. The load flow calculation was performed with ETAP 5.5 using different loading patterns generated randomly. After training of ANN, it was validated with feeding real measured value as input. The real time measured value taken from test systems was simulated in PSCAD by placing meter at the identified location of PSO. The measured value derived from PSCAD was not directly suitable to be taken as input to the estimator because the simulated data is error free. To feed noise input to the estimator the measured values with added percentage error to the respective buses were considered as testing input to the estimator.

The modified swarm tuned ANN back propagated algorithm was tested with IEEE 202 bus system and compared with ANN and WLSE to check the efficiency of the proposed algorithm. The error accuracy graph is plotted in Fig. 7, which represents the error with respect to bus number where RTU is installed for collecting real time data considering 2% error. The unmeasured bus voltage can also be estimated by algorithm excluded from this paper.

Figure 8 gives the estimated voltage with respect to different measurement error at respective buses for proposed Swarm tuned ANN compared with self tuned ANN and WLSE. For example the actual measured value at bus number 60 is 97.312% voltage magnitude. The 20% increase noise is created at bus number 60 with a measured value of 117.32% voltage magnitude. The estimated % voltage magnitude by ANN is 96.32%, WLSE is 95.32% and the proposed algorithm is 97.324%, which is almost intimately close to the measured value. Let us take another example; the measured value at bus number 175 is 93.24%. The 10% decrease in noise is created at bus number 175 with a measured value of 83.84%. The estimated % voltage magnitude by ANN is 92.84%, WLSE is 91.84% and the proposed algorithm is 93.84%, which is almost close to the measured value. From the above description and Fig. 8, it is crystal clear that, the proposed estimator is best suited for all network to achieve precise results. The efficiency of the proposed algorithm when compared with those in Refs. [23-25] is high enough to be titled the best estimator to receive the best estimate value with minimum error and computation time.

The modified proposed swarm tuned ANN back propagated algorithm tested with TNEB 140 bus system is compared with ANN and WLSE to check the efficiency of the proposed algorithm. The error accuracy graph is plotted in Fig. 9, which represents the error with respect to bus number where RTU is placed to collect the real time data considering 2% measurement error. The unmeasured bus voltage can also be estimated by algorithm excluded from this paper.

Figure 10 represents the estimated voltage with respect to different measurement error at respective buses for proposed Swarm tuned ANN compared with ANN and WLSE. For example the actual measured value at bus number 19 is 98.82% voltage magnitude. The 10% increase noise is created at bus number 19 with a measured value of 89.98% voltage magnitude. The estimated % voltage magnitude by ANN is 97.97%, WLSE is 95.97% and the proposed algorithm is 99.01%, which is almost near to the measured value. Let us take another example; the measured value at bus number 119 is 96.24%. The 5% decrease in noise is created at bus number 119 with a measured value of 91.24%. The estimated voltage magnitude by ANN is 95.75%, WLSE is 95.074% and the proposed algorithm is 96.074%, which is near to the measured value. From the above description and Fig. 10, it is cleared that, the proposed estimator is best suited for all networks to achieve accuracy results. The efficiency of the proposed algorithm when compared with those in Refs. [7,20,29] is the best estimated value with minimum error and computation time.

Conclusions

The formulated algorithm to identify meter location by particle swarm optimization and estimating bus voltage magnitude by swarm tuned ANN estimator were tested with IEEE and TNEB systems. The real time measured value was monitored through proposed identified location nodes. The noise level in the meter were selected randomly±20% at the selected nodes. The executed outputs of the proposed algorithms wre compared with mathematical and other intelligent algorithms to check the superiority of the proposed algorithm. The simulated results proves that the swarm tuned ANN estimator is best suited for accurate estimation of voltage with different noise levels. The future scope of this work is to include the impact of DG penetration in distribution system for meter placement and state estimation.

References

[1] Scheppe F C, Wildes J, Rom D B. Power system state estimation, Part 1-3. IEEE Transactions on Power Apparatus and Systems, 1970, PAS-89(1): 120–135

[2] Ramarao N, Krishna G M, Mohammed M. A new algorithm for power system state estimation. Proceedings of the IEEE, 1982, 70(10): 75–81

[3] Ramesh L, Chowdhury S P, Chowdhury S, Crossley P A. Electrical power system state estimation meter placement—A comparative survey report. Electrical Power Components and Systems. 2008, 36(10): 1115–1129

[4] Baran M E, Zhu J, Kelley A W. Meter placement for real time monitoring of distribution feeders. IEEE Transactions on Power Systems, 1996, 11(1): 332–337

[5] Souza J C S, Filho M B C, Schilling E M M. Planning metering system for power distribution systems monitoring. In: IEEE Bologna Power Tech Conference, Bologna, Italy, 2003, 107–111

[6] Wang H, Schulz N N. A revised branch current-based distribution system state estimation algorithm and meter placement impact. IEEE Transactions on Power Systems, 2004, 19(1): 207–213

[7] Das B. Rule based algorithm for meter placement and ANN based bus voltage estimation in radial power distribution system. Electrical Power Components and Systems, 2005, 33(4): 449–462

[8] Shafiu A, Jenkins N, Strbac G. Measurement location for state estimation of distribution networks with generation. IEEE Proceedings—Generation, Transmission and Distribution, 2005, 152(2): 240–246

[9] Muscas C, Pilo F, Pisano G, Sulis S. Optimal allocation of multichannel measurement devices for distribution state estimation. IEEE Transactions on Instrumentation and Measurement, 2009, 58(6): 1929–1937

[10] Cecchi V, Yang X, Miu K, Nwankpa C O. Instrumentation and measurement of a power distribution system laboratory for meter placement and network reconfiguration studies. IEEE Transactions on Instrumentation and Measurement, 2007, 56(4): 1224–1230

[11] Bignucolo F, Caldon R. Optimizing the voltage measurements location for management of active distribution network. In: University Power Engineering Conference, UK, 2007, 960–964

[12] Ramesh L, Chowdhury S P, Chowdhury S, Natarajan A A. Planning optimal intelligent metering for distribution system monitoring and control. In: IEEE INDICON 2008, Kanpur, 2008, 218–222

[13] Ramesh L, Chowdhury S P, Chowdhury S, Gaunt C T. A literature review on optimum meter placement algorithms for distribution state estimation. In: International Conference IASTED Euro PES2009, Crete, 2009

[14] Singh R, Pal B C, Vinter R B. Measurement placement in distribution system state estimation. IEEE Transactions on Power Systems, 2009, 24(2): 668–675

[15] Muscas C, Pilo F, Pisano G, Sulis S. Optimal placement of measurement devices in electric distribution systems. In: IMTC 2006—Instrumentation and Measurement Technology Conference, Sorrento, 2006, 1873–1878

[16] Ramesh L, Chowdhury S P, Chowdhury S, Natarajan A A, Song Y H, Goswami P K. Distributed state estimation technique for active distribution networks. In: IEEE Proceeding of UPEC 2007, 2007, 861–866

[17] Gelagaev R, Vermeyen P, Driesen J. State estimation in distribution grids. In: 13th IEEE International Conference on Harmonics and Quality of Power, Wollongong, 2008, 1–6

[18] Singh R, Pal B C, Jabr R A. Choice of estimator for distribution system state estimation. In: IET Generation Transmission Distribution, 2009, 3(7): 666–678

[19] Baran M, McDermott T E. Distribution system state estimation using AMI data. In: Proceedings of IEEE Power System Conference and Exposition, 2009, 1–3

[20] Singh R, Pal B C, Jabr R A. Distribution system state estimation through Gaussian mixture model of the load as pseudo-measurement. IET Generation Transmission Distribution, 2010, 4(1): 50–59

[21] de Souza J C S, Filho M B C, Schilling M, de Capdeville C. Optimal metering systems for monitoring power networks under multiple topological scenarios. IEEE Transactions on Power Systems, 2005, 20(4): 1700–1708

[22] Xu B, Abur A.Optimal Placement of Phasor Measurement Units for State Estimation. Final Project Report PSERC Publication 05-58, Texas A&M University, 2005

[23] Ranjbar H M, Amraee A M, Shirani T. Optimal placement of phasor measurement units: Particle swarm optimization approach. In: Proceedings of International Conference on Intelligent Systems Applications to Power Systems, 2007, (1): 1–6

[24] Moradi A, Fotuhi M. Optimal switch placement in distribution system using trinary particle swarm optimization algorithm. IEEE Transactions on Power Delivery, 2008, 23(1): 271–279

[25] Engelbrecht P. Computational Intelligence: An Introduction. 2nd ed. England: John Wiley and Sons Ltd., 2007

[26] Bernieri A, Betta G, Liguiri C, Lusi A. Neural network and pseudo-measurements for real time monitoring of distribution systems. IEEE Transactions on Instrumentation and Measurement, 1996, 45(2): 645–650

[27] Manitsus E, Singh R, Pal B, Strbac G.Modelling of pseudo-measurements for distribution system state estimation. IET CIRED Seminar 2008, Frankfurt, 2008, Paper No. 0018

[28] Wan J, Miu K N. Weighted least squares methods for load estimation in distribution networks. IEEE Transactions on Power Systems, 2003, 18(4): 1338–1345

[29] Naka S, Genji T, Yura T, Fukuyama Y. A hybrid particle swarm optimization for distribution state estimation. IEEE Transactions on Power Systems, 2003, 18(1): 60–68

[30] Kennedy J, Eberhart R. Particle swarm optimization. In: Proceedings of IEEE International Conference on Neural Networks, Perth, 1995, 4: 1942–1948

[31] Su T, Jhang J, Hou C. A hybrid artificial neural networks and particle swarm optimization for function approximation. International Journal of Innovative Computing, Information and Control, 2008, 4(9): 2363–2374

[32] Gao L, Zhou C, Gao H, Shi Y.Combining PSO and neural network for diagnosis of unexplained syncope. Lecturer Notes in Computer Science, 2006, 4115: 174–181

[33] Zilouchian A, Jamshidi M. Intelligent Control Systems Using Soft Computing Methodologies. New York: CRC Press, 2001

[34] Knight A. Basics of MATLAB and Beyond. London: CRC Press LLC, 2000

[35] Lyshevski S E. Engineering and Scientific Computations Using MATLAB. Hoboken: John Wiley & Sons Inc., 2003

[36] Ramesh L, Chowdhury S P, Chowdhury S, Natarajan A A. Intelligent meter placement approach for power distribution system. Journal of Electrical Engineering, 2009, 9(3): 103–110

[37] Muscas C, Pilo F, Pisano G, Sulis S. Optimal placement of measurement devices in electrical distribution systems. In: Proceedings of IEEE Instrumentation and Measurement Technology Conference, Sorrento, 2006, 1873–1878

[38] Ramesh L, Chowdhury S P, Chowdhury S, Gaunt C T. Distribution system intelligent state estimation through minimal meter placement. In: Proceedings of IASTED International Conference Asia PES 2009, Beijing, 2009, 658–114

RIGHTS & PERMISSIONS

Higher Education Press and Springer-Verlag Berlin Heidelberg