الفهرس 
IntroductionOnly 14 pages are availabe for public view
 |  | from | 178 |  |  |
| | | from | 178 | | |
Abstract
This thesis proposes an effective fault location technique for a multi-terminal DC Microgrid. Zonal fault detection is first identified via resistance estimation at both cable terminals via intelligent protection devices by capturing voltages and currents at each side. Consequently, the polarities of the estimated front-end and remote-end resistances are compared by their respective intelligent devices to examine the fault. Accordingly, to decide on a cable fault occurrence, the two intelligent devices must declare negative polarities. Moreover, junction faults can be identified via the opposite polarities of voltages across two small inductors inserted around each junction. Once a fault is detected, a proposed Gaussian Process Regression-based (GPR) prediction model for fault location is initiated to estimate the fault distance. GPR is well-known for superior accuracy with the lowest possible error rate, which is also compared to other prediction techniques. To predict the fault location via GPR, the regression space is identified through three featured variables, i.e., voltage, current, and Rate of Change of Current (ROCOC) at the source-side terminal only. Additionally, the predicted fault distance for various fault inceptions, which were categorized based on fault location, and resistance, are linked to the featured variables using training data points. The accuracy of the prediction is evaluated using the Root Mean Squared Error (RMSE). The Rational Quadratic (RQ) related to the GPR model has proved its ability to locate the fault distance precisely. Another fault distance estimation technique for multi-terminal DC microgrids is introduced based on the use of Artificial Neural Networks (ANNs) trained with backpropagation algorithms to accurately locate fault distances. Three different network structures are developed and trained to handle various fault scenarios, including different fault resistances and the presence of noise. The results demonstrate high precision in fault distance estimation, with two of the structures achieving low error rates of 0.3% for the source side and 0.6% for the load side. The third structure incorporates input variables from both sides, leading to even more accurate predictions, with an error rate of less than 0.15% for both terminals. The effectiveness of the proposed approach is evaluated through time-domain investigations, considering changes in fault location and resistance for close-in, junction, and remote-end cable faults. Furthermore, an experimental setup involving a DC microgrid powered by a PV system and battery banks is implemented, considering both radial and ring configurations. The significance of the proposed fault location approach is assessed by time-domain investigations against changes in fault location and resistance for close-in, junction, and remote-end cable faults. Also, an experimental setup is executed via a DC-MG powered by PV system and battery banks for radial as well as ring configurations.