Fault location in unbalanced distribution system including distributed generation units using multi-layer feed forward neural network

• Main Author: Farzan, Payam
• Format: Thesis
• Language: English
• Published: 2014
• Online Access: http://psasir.upm.edu.my/id/eprint/64305/1/FK%202014%20120IR.pdf; http://psasir.upm.edu.my/id/eprint/64305/
• Institution: Universiti Putra Malaysia

Locating a fault in a distribution system has always been a critical issue for electrical utilities. Fast and accurate determination of fault location results in speeding up the restoration operation and preventing waste of generated electricity in the form of undistributed energy. Fault location finding process in distribution network is totally different based on the application of developed algorithms for the transmission lines due to characteristics of distribution system. On the other hand, Distribution Generation (DG) units are increasingly being added nowadays to the distribution network. Considering the imposed impacts of these units on the distribution networks the fault location operation has even become rather complicated than before. This thesis presents a fault location algorithm based on the recording of Short Circuit Power (S/C.P) and Short Circuit Current (S/C.C) values at the source bus of unbalanced radial simulated distribution networks including DG units. The recorded values are gathered in separated datasets to be evaluated by the designed Multi-Layer Feed Forwarded Neural Networks (ML-FFNN) and the fault distances from the source are estimated accordingly. Two radial unbalanced distribution networks are considered to implement the proposed algorithm ; IEEE 34 bus test feeder as a large scale network with the maximum length of around 60 km and a local 15 bus distribution network as a real network with several laterals. Three fault types; Three Lines (LLL) Line to Line (LL) and Single Line to Ground (SLG) are applied in different locations of distribution systems and the values of S/C.P and S/C.C with their corresponding fault distances are recorded simultaneously. The designed ML-FFNN using the three different datasets; S/C.P,S/C.C and the joined S/C.P and S/C.C, estimates the locations of faults. Finally, the estimated locations are compared with the real fault locations to calculate the difference percentage. It is explained that the estimated fault locations via S/C.P dataset are rather accurate than using S/C.C dataset and the most precise estimations belong to the joined S/C.C and S/C.P dataset for all the three fault types. Furthermore, it is indicated that the designed fault locator system is able to preserve the accuracy of estimations in presence of DG units in the distribution network using all the three datasets.