Submit Manuscript  

Article Details


Practical Implementation and Testing of RNN Based Synchronous Generator Internal-Fault Protection

[ Vol. 12 , Issue. 2 ]

Author(s):

Mohammed Ahmed Saeed* and Magdi El-Saadawi   Pages 181 - 189 ( 9 )

Abstract:


Background: Differential relay is normally used to detect faults in Synchronous Generator (SG) stator windings. Nevertheless, detection of ground fault depends on the generator grounding type. For high impedance grounding, the ground faults near the neutral terminal of the stator windings are not detectable by the differential relay. So, the ability to identify the internal fault of SG is a very important task for stable and safe operation of power systems.

Methods: Accurate algorithms for fault detection and classification based on Recurrent Neural Network (RNN) are suggested in this paper. RNNs are trained using different data available from SG MATLAB/ SIMULINK model. Simulation of different fault scenarios based on LabVIEWTM program is discussed. The studied fault scenarios include; fault type, location, resistance and fault inception angle. The RNN based algorithm is experimentally tested using an actual SG. Practical design and implementation of the proposed fault detector and classifier are presented. The hardware system is designed and built to acquire the currents at both ends of SG terminals.

Results: The presented results confirm the effectiveness of the proposed algorithm to detect minor ground faults near the neutral terminal (less than 5% of stator winding).

Conclusion: The experimental analysis shows that the proposed RNN detects and classifies the internal faults correctly, fastly and remain stable after the faults occur.

Keywords:

Recurrent Neural Network (RNN), synchronous generator protection, internal fault, fault detector, fault classifier, RNN.

Affiliation:

Faculty of Engineering, Department of Electrical Engineering, Mansoura University, Mansoura, Faculty of Engineering, Department of Electrical Engineering, Mansoura University, Mansoura

Graphical Abstract:



Read Full-Text article