Abstract:
Due to the widespread use of sensitive loads, power quality has recently became an
important issue in modern power systems. Voltage sags/swells, harmonics, voltage
imbalances, and other power quality concerns are described as any deviation in current,
voltage, or frequency that causes substantial economic losses and inconveniences to
customers. Custom power devices are an effective solution to enhance the quality of the
power supplied to the power distribution system. The series-connected Dynamic Voltage
Restorer (DVR) is one of the effective solutions to mitigate power quality problems in the
distribution system. In this thesis, the Artificial Neural Network (ANN) controlled DVR is
designed and the performance of the sensitive load connected system is investigated with a
conventional Proportional Integral (PI) controller. The Levenberg Marquardt (LM)
backpropagation algorithm is used to implement the control scheme of the Voltage Source
Inverter (VSI). Using data from the PI controller, the ANN is trained offline. The proposed
ANN-based DVR strategy was tested with a replicated model of a Hamaressa oil factory
distribution feeder, Harar, Ethiopia, by simulating in MATLAB/Simulink to show the
effectiveness of smoothing the voltage sag/swell/imbalance that occurred due to fault and
mitigation of harmonic distortion. The system’s response to load voltage is evaluated for PI based and ANN-based DVR scenarios with a maximum, minimum, and average loading
conditions. Simulation results showed that the proposed strategy effectively mitigated the
voltage sags/swells/imbalances, and reduced the load voltage Total Harmonic Distortion
(THD) to the maximum acceptable IEEE standard 519 of 1992 for harmonic distortion. The
comparison analysis of the PI and ANN controllers is also presented. The results show that
the ANN-based DVR outperforms that of the PI-based controller, which obtained a load
voltage of 94.9% and a THD of 5.04%. The ANN-based DVR achieved a load voltage of
99.5% and a THD of 1.65%