Detect, Classify, and Locate Faults in DC Microgrids Based on Support Vector Machines and Bagged Trees in the Machine Learning Approach

Abstract

The DC microgrids possess numerous pros, including enhanced reliability, increased efficiency, and a less complicated control system. Further, they provide a simplified system that facilitates the incorporation of renewable energy sources (RES), battery storage systems, and DC loads. DC microgrids improve resource coordination and utilization, thus offering a potential alternative to modern energy systems. DC power systems have unique features that make protecting DC microgrids from different types of faults very hard. These include large DC capacitors, low-impedance DC cables, no natural zero-crossing points, and significant transient current and voltage changes that happen very quickly. Also, solid-to-ground faults could result in a rapid increase in DC fault current. Therefore, a cost-effective and reliable system protection mechanism capable of detecting, locating, and isolating faults is crucial to preventing DC microgrids from experiencing power outages and failures. This paper presents a machine-learning-based protection approach for DC microgrids. The proposed methodology relies solely on measuring the current passing through the positive terminal at bus_1 in the modified IEEE 14-bus configuration. During the measurement, the DC microgrid encountered several fault scenarios. The gathered data is analyzed to train a supervised machine-learning method that uses medium-gaussian support vector machines and bagged tree classification algorithms. The effectiveness of this method was evaluated by conducting tests on a particular subset of the collected data using the trained model. The proposed protection technique was verified using MATLAB/Simulink software under several pole-pole (P-P) and pole-ground (P-G) fault conditions. The simulation results demonstrate that the proposed protection approach is practical and reliable across all fault scenarios. It is highly accurate at identifying and detecting precise fault locations without any errors.

Keywords

• LVDC microgrid
• protection
• solid fault to ground
• fault location
• fault localization
• machine learning