Fault diagnosis in unbalanced radial distribution networks using generalised regression neural network
Fault location includes the determination of the physical location of the fault. Nowadays, about 80% of interruptions are caused by faults in distribution networks and the application of fault location algorithms developed for transmission system is not an easy task due to the topology and operating...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English |
| Published: |
2011
|
| Subjects: | |
| Online Access: | http://psasir.upm.edu.my/id/eprint/41792/1/FK%202011%2017R.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Fault location includes the determination of the physical location of the fault. Nowadays, about 80% of interruptions are caused by faults in distribution networks and the application of fault location algorithms developed for transmission system is not an easy task due to the topology and operating principles of the distribution networks. This thesis describes the technique of Probabilistic Neural Network (PNN) for fault type classification and Generalised Regression Neural Network (GRNN) for estimating the fault location. The results were compared with radial basic function neural network (RBFN) and feed forward neural network (FFNN).
The artificial intelligence (AI)-based fault locator has been implemented on a typical IEEE 13 node test feeder as short feeder with the feeder’s nominal voltage is 4.16 kV. It is radial, unbalanced and includes both overhead line and underground cable and the 76-bus radial distribution system as a long feeder with two long main feeders 63/20 kV and 76 buses 20 kV. The neural network used only the voltage and current measurements obtained at the substation. The training patterns used to train the ANN model for fault location in radial distribution system were obtained by short circuit analysis under various fault conditions and fault impedances. To achieve this goal, the initial or pre-fault condition of the system has to be computed. Using the proposed method, less learning time of PNN is required for classification. The GRNN results show the effectiveness of the proposed method with good accuracy, as the fault point location determination is very close to the actual point with acceptable convergence time and accuracy. |
|---|