System Disturbance Classification Model for a Low Voltage Distribution Network

There is a growing interest in power quality issues due to the continued developments in the power grid structure, i.e. its generation composites that now include prosumers, and high integration of renewables, transmission and distribution modes are now bi-directional and the utilization of power has loads that are linear and nonlinear with demand response capability. Some power quality disturbance events have the potential to damage sensitive equipment such as programmable logic controllers, computers, and drive circuits leading to in-creases in downtime and company losses. Therefore, the need to continuously monitor and evaluate the system operating states and provide awareness to the system operators. The research aimed to develop a model trained to monitor and classify the operating system conditions. The trained model used synthetic data generated from power acceptability limits published in the IEEE and IEC standards. The training dataset consisted of 244 features, and 15 power quality event classifications each with 504 random samples. The MATLAB classification learner APP was used to train 29 classifier models by supervised learning. The trained models were tested on actual PQ metered datasets from The Copperbelt University 11 kV network. It was observed that only ten (10) models had a validation accuracy of above 90% on the training dataset and that the prediction of the unclassified events was not consistent amongst the models. The model that displayed promising performance on the training dataset (99.95% VA), test dataset (99.87% TA) and the measured PQ meter data (77.5% PA) is the Ensemble Bagged Tree model.