مشخصات پژوهش

Predicting defects and knowing the network conditions are important issues in distribution system operation. A comprehensive defect warning system considering different internal and external affecting parameters could help the operator to better manage and protect the network. A method based on ensemble learning is presented in this paper to determine the system’s status and implement a warning scheme for defects. The employed learning method is a stacking-based approach that includes two main layers. In the structure of the proposed method, the first level consists of support vector machine, random forest, and XGBoost classifiers as basic models. Also, in the next layer, referred to as the meta-classifier, a support vector machine is used as the final classifier. To identify the states of the system, voltages and currents in certain buses have been used as the system’s parameters. Weather-related effects are the other crucial factors that affect system conditions, which are included in the models. Four distinct modes are defined to specify different states of the system that actually represent the warning levels. The proposed models are evaluated using the extracted data from a modified IEEE 123-bus test system, including PVs. The obtained results are compared with several individual models, and the effect of noise on the input data has also been considered. The evaluations show highly appropriate outcomes of the proposed model in predicting different system states and achieving better results than individual methods.