Abstract:
There are numerous advantages of distributed generation systems over conventional generation systems. However, protection has always been one major challenge in a distributed generation system. From the protection's point of view, a distributed generator requires special attention on account of stability loss, failure re-closure, fluctuations in voltage, etc. The situation becomes even more challenging for a short circuit fault. And thereby it becomes substantially important to exactly identify the location & type of a fault without delay, particularly for a small distributed generation system, which otherwise impacts the operation of the system. Several techniques like traveling wave methodology, impedance-based, support vector machine (SVM), fuzzy logic, and genetic algorithm (GA) had been discussed in past to identify the type and location of a fault. However, the accuracy of all these methods has always been a major issue while incorporating them into a real system. The methodology proposed here uses an artificial neural network-based structure to identify a particular fault which can be trained with real fault and steady-state data to identify the location and type of a fault in the distributed generation system. An elementary system containing two distributed generators and a utility grid has been considered for data recording purposes. Firstly, the training of the system through the recorded simulation data is carried out followed by the validation and testing through an artificial neural network tool. After successful training and validation, the same data is tested for any given set of conditions. The modeling of the test system has been carried out on Simulink itself. The overall result shows an unprecedented zero percent error in identifying the type of fault and location.