چكيده به لاتين
Protection is the main challenge of microgrid operation and recent research works have demonstrated that conventional protection methods have shortcomings in solving microgrid protection issues. This thesis presents a new microgrid protection method based on data mining of voltage and current signals to construct feature vectors for fast fault detection and accurate faulted section identification of all fault types. These vectors are pre-processed using feature selection methods and discretisation technique. They are input into classifiers to determine the best feature vector. The simulation results show that there is a good relationship between selected features and network parameters (voltage and current) during fault events.
There are other non-fault events in a microgrid operation, which have same behavior alike faults. These events include motor starting, heavy load switching, capacitor exiting, and transformer energisation. This thesis proposes a voltage-based protection scheme using the extracted features from voltage sags to discriminate between faults and normal events. These features can identify faults regardless of microgrid structure, type, capacity, and location of distributed energy resources (DERs).
This thesis also proposes a fault classification method using series combined classifiers, which analyses fault current signals in modal domain and provides selective phase tripping. The simulation results demonstrate that the features extracted based on symmetrical components and Clarke transformations carry the most information for determining fault type and identifying the faulted phase(s), respectively.
Finally, this thesis presents data mining-based microgrid protection schemes, which satisfy important protection requirements such as speed and accuracy. The proposed schemes can be implemented in radial and meshed microgrids with islanded or grid connected modes of operation. The schemes consist of three protection functions including fault detection, faulted section identification, and fault classification. The performance of the schemes have been validated by numerous simulation results.
