Abstract:
The busbar, where all the lines converge, is very critical element of a power system. Faults on busbar can lead to disturbances and major shut down of power system. On other hand, the unwanted isolation of busbar leads to a major power loss in the system. Therefore, in order to prevent hazard and major power loss to the power system, effective and reliable protection schemes for busbars needs to be developed.
This thesis presents two schemes for fault zone identification for busbar. The first scheme is based on the Random Forest Classifier while the second is based on dq transform. The schemes depend on extracting features from the measured data of current signals of all the bays (lines) connected to a busbar. Validation of the scheme has been carried out by modelling an existing 400 kV Indian busbar system using PSCAD/EMTDC software package. A wide variety of simulation cases (33,600) consisting of internal and external faults have been generated by varying the system and fault parameters. By giving post fault data of one cycle duration of all the bay currents as input the algorithm based on RF classifier, and using 30% data for training and 70% for testing, an accuracy of more than 98% has been obtained.
Furthermore, a scheme based on dq transform has been developed. The id and iq components of all the line currents are calculated and summed up. The post fault data samples of one cycle are added to find out 𝑖𝑑𝑆𝑂𝑆 (Sum of Samples of id components) and 𝑖𝑞𝑆𝑂𝑆 (Sum of Samples of iq components). If both these values are less than the threshold, it is a case of external fault and if it is more than a threshold, it is a case of internal fault. The overall accuracy obtained with this scheme is of the order of 98%.
