DISTRIBUTION FEEDER LOSS COMPUTATION BY ARTIFICIAL NEURAL NETWORK

DISTRIBUTION FEEDER LOSS COMPUTATION BY ARTIFICIAL

NEURAL

NETWORK

S.W. Kau, M.Y. Cho, Member, E E E

Department of Electrical Engineering National Kaohsiung Institute of Technology

Kaohsiung, Taiwan, ROC

ABSTRACT: This paper proposes

an

artificial neural network (ANN) based feeder loss calculation model for distribution system analysis. In this paper, the functional-link network model is examined to form the artificial neural network architecture to derive the various loss calculation models for feeders with m e r e n t configuration. Such artificial neural network is a feedforward network that uses standard back-propagation algorithm to adjust weights on the connection path between any two processing elements (PES). Feeder daily load curve on each season are derived by field test data. Three-phase load flow program is executed to create the training sets with exact loss calculation results. A sensitivity analysis is executed to determine the key factors inchded power factor, feeder loading, primary conductors, secondary conductors, and transformer capacity as the variables for components located at input layer.

By d c i a l neural network with the pattern recognition ability, this study has developed seasonal and yearly loss calculation models for overhead and underground feeder configuration. Two practical feeders with both overhead and underground configuration in Taiwan Power Company (TPC or Taipower) distribution system are selected for computer simulation to demonstrate the effectiveness and accuracy of the proposed models. As comparing with models derived by the conventional regression technique, results indicate that the proposed models provide more efficient tool to District engineer for feeder loss calculation.

1. INTRODUCTION

Power company usually assess the operating efficiency with the amount of real power loss introduced ftom components

in

power system. For a power system, the distribution system loss has become a concerned topic because of the growth of load and the wide area it covers. However, the conventional loss analysis that applies the detail system modeling is difficult and impractical to be performed since voluminous data are involved. The primary conductors, secondary conductors, and transformers normally contribute most of the real power loss for a distribution system. Besides, the variation of feeder characteristics such as load imbalance and phase voltage imbalance etc. will introduce the change of load demand and system loss.

The relative methodologies for system loss calculation have been applied to distribution system analysis successllly in the last two decades. These loss calculation technologies consist of percent loading method, simplified feeder model, load duration c w e s derivation, and load window etc. [l-61. Currently, the loss calculation method taken by the engineer in the TPC District is as follow. Two loss calculation models, the exact and simplified feeder loss models, [7] are utilized for distribution feeder loss analysis. The exact feeder loss model is mainly designed to evaluate the feeder with

heavy loading and longer conductor length. It provides loss components information such as primary conductor loss, transformer coDDer loss. transformer core loss. and secondarv conductor loss as well as total amount of feeder loss to engineer for complete loss analysis. On the other hand, the simplised feeder loss model is designed to rapidly obtain real power loss of feeder with light loading and shorter conductor length. Since the AM/FM (automatic mappmg&cility management) system is not available from TPC District, the statistical data for feeder is manually extracted from maps with U600 scaling. By using exact feeder loss model, District engineer normally spent 5-7 working days to prepare required data of one feeder for computer simulation . Besides, the time spent of data collection for computer simulation by using simplified feeder loss model needs only several working hours. By the experience obtained from District engineer, most of feeders have light loading and shorter conductor length. It means that they introduce lower real power loss below the average value. Therefore, the loss calculation for most of feeder are accomplished by the simplified feeder loss modeL

In TPC distribution system, a feeder may serve the mixture load of various types of customers. To perform the computer simulation, the feeder load curve at each season is obtained by integrating the typical load pattern of each type of customer [8]. By inspecting the simplilied feeder loss model derived &om conventional regression technique, the real power loss introduced is varied according to the load proflle for the feeder with same configuration. Thus, in the long run there exist some loss patterns that correspond to load prone at various seasons. Besides, the accuracy of feeder loss resulted by simplified feeder model is not satisfactory [7] due to the drawback of narrow effective data range, inefficient data manipulation, etc. These observation motivate

d e

use of ANN for feeder loss calculation.

In this paper, the ANN are used for fast pattern recognition and regression of the feeder loss model so that the effort for feeder loss calculation can be reduced. The topology of the proposed ANN, the selection of the input and output variables, training set structure, and training method are discussed in this paper. Functional link back- propagation network with sine version is selected for our ANN model because it is effective in the area of regression applications. Several ANN models corresponding to seasonal and yearly feeder loss models are proposed to support distribution system analysis.

2. FEEDER LOSS AND

SENSITIVITY

ANALYSIS 2.1 Derivation of Feeder Load Curve

In order to support ANN model development and to manipulate the feeder loss more practically, the typical feeder daily load curve for each season is derive by the field test. The power consumption recorders are installed at substation to collect the energy consumption of the study feeders during each 5 minute time interval over one year. Figs. 1 and 2 show the typical da;ly load curve of

feeder MJ66 and L132 located at TPC Kaohsiung District distribution system The load characteristic of these two feeders are quke Herent. Feeder MJ66 serves the mixture load of residential and commercial customers while feeder L132 serves both the residential and small industrial customers.

Figs. 7 and 8 represent the daily loss patterns of feeders MI66 and L132 respectively in summer. The loss obtained by exact feeder loss model is used as reference. In these two figures, except for loss at few time intervals, the loss obtained by ANN models are more close to the loss calculated by conventional regression technique at all time intervals. Finally, in order to support more efficient loss model to District engineer, this paper is tried to derive the yearly feeder loss model. By the composite load curve and the training set covered over one year round, the yearly feeder loss models for feeders MJ66 and L132 are then obtained. Since more training sets are presented, more leaming numbers as well as training time are required to complete training procedure. Results obtained fiom recalling process show that the error introduced is about 0.34%. Fig. 9 shows that the loss obtained by yearly ANN model is nearly close to the desired loss calculated by exact loss model.

1

l e x a c t m o d e l B A N N S m o d e l -regression m o d e l l 2 0 0 150

g

1 0 0 I -1 5 0 0 1 3 5 7 9 1 1 1 3 1 5 1 7 1 9 2 1 2 3 T i m e

Fig. 7 DaiIy loss pattem of feeder MJ66 in summer

l e x a c t m o d e l I A N N s m o d e l I f e g r e s s i o n m o d e l 200 150

2

100 50 0 1 3 5 7 9 11 1 3 15 1 7 1 9 2 1 2 3 T i m e

Fig. 8 Daily loss pattem of feeder L132 in summer

I

I

e x a c t m od e l I A N N s m ode1

I

250 r 200

2

150

g

100 -3 50 0 1 3 5 7 9 11 13 15 17 19 21 2 3 Time

Fig. 9 Loss results of feeder MJ66 by using yearly

ANN

model

In TPC distribution system, the linkage between customers and distribution transformers is not available except for some key customers. This situation increases the diflicult of computer simulation executed by conventional methods. The ANN approach does not need this linkage information and works for loss calculation regularly. After the ANN model has been trained, the recalling process is so fast that the time for loss calculation can be saved. Besides, the advantage of the A N N s approach is that it can be continuously improved by retaining the network architecture with new data as the feeder reconfigures. For distribution system

with

hundreds of feeders, it is diftlcult to solve the loss by using detail loss model because of voluminous data need to be prepared. Therefore, the proposed ANN models is developed to solve loss efficiently and to save the time for data collection.

5. CONCLUSION

An artificial neural network based feeder loss calculation model for distribution system analysis is presented in

this

paper. The functional-link network model is selected to build various loss calculation models for feeders with Werent configuration and load characteristic. Such artificial neural network is a feed-forward network that uses standard back-propagation algorithm to adjust weights on the connection path between any two PES. Feeder daily load pattems and statistical data are used to establish ANN models. A sensitivity analysis for feeder loss calculation is executed to determine the key factors affected strongly the loss as the variable signals for input layer.

By artificial neural network with the pattern recognition ab*, this study has developed seasonal, and yearly feeder loss calculation models for overhead and underground feeder configuration. Simulation results show that the proposed ANN models are superior to conventional regression method. Experience has shown that the presented approach allows the District engineer to calculate loss efficiently, save time spent for data collection.

6. REFERENCE

D.I.H. Sun, et al. "Calculation of Energy Losses in a Distribution System", E E E Trans. on Power Apparatus and System, Vol., PAS-99, No. 4, 1980, pp. 1347-1356.

Nelson E. Chang, "Determination of Primary Feeder Losses", EEE Trans. on Power Apparatus and System, Vol., PAS-87, N.R Schul- "Distribution Primary Feeder Losses", E E E Trans. on Power Apparatus and System, Vol., PAS-97, No. 2; Insulated Conductors Committee, "Loss Evaluation for Underground Transmission and Distriiution Cable System", IEEE Trans. on Power Delivery, Vol. 5 , No. 4, 1990, pp. D.L. Flaten, "Distribution System Losses Calculated by Percent Loading", IEEE Trans. on Power System, Vol. No. 2, N. Vempati et al., "Simplified Feeder Modeling for Load Flow Calculation", IEEE Trans. on Power System, Vol.

C.S. Chen, J.C. Hwang, M.Y. Cho, Y.W. Chen, "Development of Simplified Loss Models for Distribution System Analysis", JEEE Trans. on Power Delivery, Vol. 9, C. S. Chen, J.S. Wu, M.T. Tsay, C.T. Liu, "Application of Load Pattems and Customer Information System for Distribution System Analysis", Proceeding of the First International Conference on Power Distribution, Brad, Nov. 1990. NO. 12, 1968, pp. 1991-1994. 1978, pp. 603-609. 1652-1659. 1988, pp. 1263-1269. PWRS-2, NO. 1, 1987, pp. 168-174. NO. 3, July, 1994, pp. 1545-1551. 77

Q. Zhou, J. Davidson, A.A. Fouad, "Application of Artificial Neural Networks in Power System Security and Vulnerability Assessment", IEEE, paper No. 93 WM 183-4 PWRS. D.J. Sobajic, Y.H. Pao, "Artificial Neural-Net Based Dynamic Security Assessment for Electric Power Systems", IEEE Trans. on Power Systems, Feb. 1989, pp. 220-228.

C.N. Lu, H.T. Wu, S. Vemuri, "Neural Network Based Short Term Load Forecasting", IEEE Trans. on Power System, Vol. 8, No. 1, Feb. 1993, pp. 336-342.

C.N. Lu, M.T. Tsay, Y.J. Hwang, Y.C. Lm, "An Artificial Neural Network Based Trouble Call Analysis", IEEE, Trans. on Power Delivery, Vol. 9, No. 3, July, 1994, pp. 1663-1668. Damir Novosel, Roger L. King, "Using Artificial Neural Networks for Load Shedding to Alleviate overloaded Lines", IEEE, paper No. 93 WM 036-4 PWRD.

Hoyong Kim, Yunseok KO, Kyung-Hee Jung, "Artificial Neural-Network Based Feeder Reconfiguration for Loss Reduction in Distribution Systems", IEEE, Trans. on Power Delivery, Vol. 8, No. 3, July, 1993,pp. 1356-1366.

Jacek M. Zurada, Introduction to

Artificial

Neural Systems, Info Access Distribution Pte Ltd. 1992.

R Durbin, C. Miall, G. Mitchison (Ed), The Computing Neuron, Addison-Wesley Publishing Co.