Design of artificial neural networks for distribution feeder loss analysis

Design of artificial neural networks for distribution feeder loss analysis

Tsung-En Lee

, Chin-Ying Ho

, Chia-Hung Lin

, Meei-Song Kang

a

Department of Electrical Engineering, National Kaohsiung University of Applied Sciences, Kaohsiung, Taiwan

b

Department of Electrical Engineering, Kao Yuan University, Lu Chu, Taiwan

a r t i c l e

i n f o

Keywords:

Artificial neural network Levenberg–Marquardt algorithm Outage management system Customer information system

a b s t r a c t

To enhance the efficiency for power loss analysis of voluminous distribution feeders, ANN-based simplified power loss models with the Levenberg–Marquardt (LM) algorithm have been developed for overhead feeders and underground feeders, respectively. The three-phase load flow analysis is executed to obtain the sensitivity of feeder loss with variations in power loading, conductor length, and total capacity of distribution transformers. Through this, the data set for neural network training is prepared to derive the ANN-based simplified power loss models. The power loss of each distribution feeder can be easily derived from the key factors of hourly loading, feeder length, and transformer capacity. By integrating the power loss of all feeders, the power loss of the entire distribution system can thus be obtained to estimate the operation efficiency of the Taipower system.

Ó 2011 Elsevier Ltd. All rights reserved.

1. Introduction

The distribution system of Taiwan Power Company (Taipower) has contributed more than 50% of total system loss, posing prob-lems in the pursuit of higher system operation efficiency (Chen & Cho, 1993). It has become a critical issue for Taipower. Thus, the power loss of each distribution feeder is examined so that effective strategies, such as reactive power compensation and optimal sys-tem configuration, can be implemented (Chen, Hwang, Cho, & Chen, 1994). With more than 6000 feeders in the Taipower system, performing conventional loss analysis of distribution systems with the input data (Gustafson & Baylor, 1988; Oliveira et al., 2001) and identifying the power demand of all load buses (Chen, Kang, Hwang, & Huang, 2001; Gustafson & Baylor, 1989; Saenz, Eguia, Berasategui, Marin, & Arceluz, 2001) are both tedious and impractical.

Up to now, different methodologies such as percent loading method, simplified feeder model, load duration curve and load window (Chang, 1968; Flaten, 1988; Insulated Conductors Committee, 1990; Schultz, 1978; Sun et al., 1980) have been employed for distribution system loss analysis. Both exact and simplified loss models (Chen et al., 1994) have been considered for each service district of Taipower to solve the line loss and transformer loss for distribution feeders.

Accurate representation of distribution network topology is important for feeder loss analysis. For conventional distribution loss analysis, feeder network configuration is manually extracted

from paper maps and attributes of all distribution components are retrieved from different data files. It is very tedious and time-consuming for distribution engineers to collect and prepare the in-put data of distribution feeders for comin-puter simulation. With the implementation of the outage management system (OMS), the net-work configuration and input data for loss analysis can be gener-ated with the support of automgener-ated mapping and facility management function at Taipower (Ockwell, 2003). The topology process and connectivity trace of electric circuits are performed to identify the network configuration and all customers served by each distribution transformer.

Owing to variations in load composition and electricity con-sumption pattern of different appliances, the load characteristics of residential, commercial and industrial customers will affect the hourly loading level of each distribution transformer (Chen, Hwang, Tzeng, Huang, & Cho, 1996; McCuen, 1985). In order to solve feeder loss more accurately, the typical daily load patterns of all customer classes during each season are considered in this paper. After identifying the customers served by each distribution transformer, the power consumption of these customers is re-trieved from the database of the customer information system (CIS). The hourly power demand of each customer is then deter-mined by allocating the monthly energy consumption to each time lot according to the corresponding typical load pattern.

Artificial neural networks (ANNs) have been given much atten-tion by power engineers in the past few years. Many interesting applications of neural nets in the power field have been reported, such as short-term price forecasting in deregulated market (Areekul, Senjyu, Toyama, & Yona, 2010), electric arc furnace load model analysis (Chang, Chen, & Liu, 2010), intelligent thermographic diagnostic (Laurentys Almeida et al., 2009), short-term load 0957-4174/$ - see front matter Ó 2011 Elsevier Ltd. All rights reserved.

doi:10.1016/j.eswa.2011.05.064 ⇑Corresponding author.

E-mail address:[email protected](C.-H. Lin).

Expert Systems with Applications 38 (2011) 14838–14845

Contents lists available atScienceDirect

Expert Systems with Applications

forecasting (Bashir & El-Hawary, 2009; Fan, Chen, & Lee, 2009) and passive harmonic filters (Wu, 2007).

To derive the neural-based simplified loss models of distribu-tion feeders, the multilayer feedforward neural network has been proposed in this paper. The feeder loading, transformer capacity, and conductor length are considered as the neurons in the input layer while the feeder loss is the neuron in the output layer. To solve the weights of connectivity arcs and the bias between neu-rons of adjacent layers in the neural network, a set of training pat-terns are obtained by executing the 3  / load flow analysis to obtain power loss of distribution feeder with different scenarios of feeder loading, transformer capacity and conductor length. To accelerate the learning process, the Levenberg–Marquardt Back-Propagation (LMBP) neural learning technique is used in the ANN training process to achieve more efficient convergence rate. To demonstrate the effectiveness and accuracy of the proposed neu-ral-based distribution loss analysis, various types of distribution feeders in the Taipower system are selected for computer simula-tion to derive the ANN-based simplified feeder loss models. With feeder loss models obtained, the hourly power loss of each distri-bution feeder can be easily solved according to the type of custom-ers served, total conductor length and total capacity of distribution transformers.

The rest of the paper is organized as follows. Section2describes the information system integration and the sensitivity of feeder loss with variations in power loading, conductor length, and total capacity of distribution transformers. Section3provides the design of the artificial neural network for feeder loss analysis. Simulation results for the practical Taipower distribution feeders are provided in Section4. Discussion and conclusion are presented in Section5

to explain the effectiveness of results obtained by the proposed method.

2. Distribution system loss analysis

To derive the simplified loss models for various types of distri-bution feeders, the electrical network topology of distridistri-bution sys-tems is identified and the corresponding bus data, branch data, and customer data are retrieved from the OMS database and CIS data-base. With the connectivity of distribution transformer and cus-tomers, the hourly loading of each load bus is obtained by integrating the power profiles of all customers served. A three-phase load flow analysis is executed to solve power loss of distri-bution feeders and to create the training data set. The overall development process of distribution feeder loss modeling is illus-trated inFig. 1.

2.1. Data preparation for three-phase load flow analysis

To prepare the data files for conventional load flow analysis, a lot of effort will be required in collecting system component data such as lengths and sizes of line conductors, impedances and capacities of distribution transformers, and power consumption of all customers served. To support various applications of the dis-tribution management system, Taipower has been implemented the OMS by digitizing the geo-based distribution networks and storing the attributes of system components in the database. In this paper, the attributes of distribution components are retrieved from the database, and the topology process is performed to iden-tify the network configuration of distribution feeders. Through this, the input data files for feeder load flow analysis can be easily obtained.

For three-phase load flow analysis, the hourly loadings of distri-bution transformers are evaluated in this paper. The monthly en-ergy consumption of each customer, which has been retrieved

from the CIS of Taipower, is allocated to each study hour according to the corresponding load pattern (Chen et al., 1996). The power demand of each load bus or distribution transformer is then calcu-lated by integrating the hourly loading of all customers served. The equivalent circuits of three-phase, two-phase, and single-phase line segments are derived by considering the mutual coupling of phase conductors and grounding effect (Sun et al., 1980). In this way, the distribution system power loss is solved in a rather accu-rate manner by considering the variations in customer load behav-ior and three-phase unbalance of distribution systems.

2.2. Feeder loss sensitivity analysis

To prepare the training data set of neural networks for the sim-plified neural-based loss models of distribution feeders, the sensi-tivity analysis of feeder loss with respect to feeder loading, conductor length, and total transformer capacity is performed for different types of distribution feeders. The three-phase load flow analysis is executed to solve the power loss of test feeders by vary-ing the above factors for each study case.

2.2.1. Effect of feeder loading

Increase in feeder loading will cause larger current flow, which will result in larger line loss and transformer copper loss. Fig. 2

shows the relationship between various types of system losses and feeder loading. As can be seen, both transformer copper loss and line loss increase with the square of feeder loading, while the transformer core loss decreases with feeder loading because of larger voltage drop.

2.2.2. Effect of conductor length

Because conductor resistance increases with conductor length, a longer feeder will introduce larger feeder loss. On the other hand, the transformer core loss decreases slightly with conductor length due to larger voltage drop.Fig. 3shows the relationship between system losses and feeder length. The primary line loss is linearly proportional to feeder length.

2.2.3. Effect of transformer capacity

Fig. 4shows the relationship between system losses and trans-former capacity at a given load level. Since the transtrans-former core

Topology Process

Node reduction

CIS

Typical load patterns of customer classes

Determination of transformer hourly loading

Perform neural network training to derive the simplified loss models of distribution feeders

Load survey study

Connectivity of distribution transformers

and customers

Circuit modeling of distribution feeder

3- load flow analysis to create training data set Retrieve the attributes of

distribution components from the OMS facility database for test feeder

Fig. 1. Process of simplified feeder loss modeling.

rather consistent with that solved by the three-phase load flow analysis with the average mismatch of 2.22% and 1.22% for Feeder MB74 and Feeder LY32, respectively.

4.2. Power loss of Taipower distribution system

After deriving the ANN-based simplified power loss models for both overhead and underground distribution feeders of Taipower, the hourly power loss of each feeder is therefore obtained accord-ing to the feeder type, power loadaccord-ing, total transformer capacity, and conductor length. For all the distribution feeders of Taipower, the hourly power loadings of 2009 have been recorded in the SCA-DA database. The total capacity of distribution transformers and total conductor length of line segments have been retrieved from the OMS for each distribution feeder.

Fig. 10shows the monthly power loss and loss percentage of Taipower distribution system in 2009. In July, the distribution power loss reaches the maximum value of 504.4 MWh or 3.06% of total system generation when system peak loading occurs.

5. Conclusion

This paper has presented a systematic approach to analyzing the power loss of distribution systems. To improve the efficiency of feeder power loss analysis for distribution systems, ANN-based simplified power loss models have been derived for both overhead feeders and underground feeders, respectively. The data set for ANN training has been created by executing the three-phase load flow analysis to find the power loss of testing feeders with different scenarios of network configuration and power loading level. By retrieving the total capacity of distribution transformers and total conductor length of line segments for each feeder from the outage management system of Taipower, the power loss of all distribution feeders is solved using the derived ANN-based simplified power loss models according to the hourly power loadings. By integrating the power loss of both distribution system and transmission sys-tem, the total power loss of Taipower system in 2009 has been ob-tained. The average monthly power loss of Taipower distribution system has been founded to be 385.6 MWh, which represents 2.6% of total power generation.

References

Areekul, P., Senjyu, T., Toyama, H., & Yona, A. (2010). A hybrid ARIMA and neural network model for short-term price forecasting in deregulated market. IEEE Transactions on Power Systems, 25(1), 524–530.

Bashir, Z. A., & El-Hawary, M. E. (2009). Applying wavelets to short-term load forecasting using PSO-based neural networks. IEEE Transactions on Power Systems, 24(1).

Chang, N. E. (1968). Determination of primary feeder losses. IEEE Transactions on Power Apparatus and Systems, PAS-87(12), 1991–1994.

Chang, G. W., Chen, C. I., & Liu, Y. J. (2010). A neural-network-based method of modeling electric arc furnace load for power engineering study. IEEE Transactions on Power Systems, 25(1), 138–146.

Chang, Y. P., Low, C., & Wu, C. J. (2007). Optimal design of discrete-value passive harmonic filters using sequential neural-network approximation and orthogonal array. IEEE Transactions on Power Delivery, 22(3), 1813–1821. Chen, C. S., & Cho, M. Y. (1993). Energy loss reduction by critical switches. IEEE

Transactions on Power Delivery, 8(3), 1246–1253.

Chen, C. S., Hwang, J. C., Cho, M. Y., & Chen, Y. W. (1994). Development of simplified loss models for distribution system analysis. IEEE Transactions on Power Delivery, 9(3), 1545–1551.

Chen, C. S., Hwang, J. C., Tzeng, Y. M., Huang, C. W., & Cho, M. Y. (1996). Determination of customer load characteristics by load survey system at Taipower. IEEE Transactions on Power Delivery, 11(3), 1430–1436.

Chen, C. S., Kang, M. S., Hwang, J. C., & Huang, C. W. (2001). Temperature effect to distribution system load profiles and feeder losses. IEEE Transactions on Power Systems, 16(4), 916–921.

Fan, S., Chen, L., & Lee, W. J. (2009). Short-term load forecasting using comprehensive combination based on multimeteorological information. IEEE Transactions on Industry Applications, 45(4), 1460–1466.

Flaten, D. L. (1988). Distribution system losses calculated by percent loading. IEEE Transactions on Power Systems, 3(3), 1263–1269.

Gustafson, M. W., & Baylor, J. S. (1988). Transmission loss evaluation for electric systems. IEEE Transactions on Power Systems, 3(3), 1026–1032.

Gustafson, M. W., & Baylor, J. S. (1989). Approximating the system losses equation. IEEE Transactions on Power Systems, 4(3), 850–855.

Hagan, M. T., & Menhaj, M. B. (1994). Training feedforward networks with the Marquardt algorithm. IEEE Transactions on Neural Networks, 5(6), 989–993. Haykin, S. (1994). Neural network: A comprehensive foundation. Englewood Cliffs, NJ:

Prentice-Hall.

Insulated Conductors Committee (1990). Loss evaluation for underground transmission and distribution cable system. IEEE Transactions on Power Delivery, 5(4), 652–1659.

Laurentys Almeida, C. A., Braga, A. P., Nascimento, S., Paiva, V., Martins, H. J. A., Torres, R., et al. (2009). Intelligent thermographic diagnostic applied to surge arresters: A new approach. IEEE Transactions on Power Delivery, 24(2), 751–757. McCuen, R. H. (1985). Statistical methods for engineers. Prentice-Hall.

Ockwell, G. L. (2003). Implementation of network reconfiguration for Taiwan Power Company. In IEEE Power Engineering Society general meeting, 13–17 July, 2003, Toronto, Ontario, Canada.

Oliveira, C. C. B., Kagan, N., Meffe, A., Jonathan, S., & Caparroz, S., & Cavaretti, J. L. (2001). A new method for the computation of technical losses in electrical power distribution systems. In Proceedings of the Institute of electrical and electronics engineers, CIRED, June 18–21, 2001. Conference Publication No. 482. 0 50 100 150 200 250 300 350 400 450 500

Jan. Feb. Mar. Apr. May Jun. Jul. Aug. Sep. Oct. Nov. Dec. Month Power loss(MWh) 2.1 2.2 2.3 2.4 2.5 2.6 2.7 Loss(%) Power loss Loss percentage

Fig. 10. Monthly power loss and loss percentage of Taipower distribution system (2009). 14844 T.-E. Lee et al. / Expert Systems with Applications 38 (2011) 14838–14845

Saenz, J. R., Eguia, P., Berasategui, J. L., Marin, J., & Arceluz, J. (2001). Allocating distribution losses to customers using distribution loss factors. In Proceedings of IEEE Porto power tech conference, September 10–13, Porto, Portugal.

Schultz, N. R. (1978). Distribution primary feeder losses. IEEE Transactions on Power Apparatus and Systems, PAS-97(2), 603–609.

Sun, D. I. H., Abe, S., Eichenberger, P., Farris, D., Shoults, R. R., & Chen, M. S. (1980). Calculation of energy losses in a distribution system. IEEE Transactions on Power Apparatus and Systems, PAS-99(4), 1347–1356.

Wilamowski, B. M., Iplikci, S., Kaynak, O., & Efe, M. O. (2001). An algorithm for fast convergence in training neural networks. In Proceedings of IEEE international joint conference on neural networks, July 2001, Washington DC, USA.