Reducing Number of Communication Devices In Microgrids Using Machine Learning

In microgrids there exists multiple buses that have a noticeable difference in either phasor voltages, phasor currents, or both between the islanded mode and the grid-connected mode. Sometimes the difference could be easily distinguished in a feature of the voltage or the current like (Power, Apparent impedance, Power factor, etc ). This paper proposes a new method for microgrid islanding detection using local measurements. Using support vector machines, the authors trained a classifier that uses phasor values at a chosen bus and detect the mode of operation for a microgrid with accuracy up to 93% using only one feature.SVM classifiers can help in reducing the number of communication devices in the microgrid needed to inform the protection devices of the mode of operation in order to change the settings for the relay to best match the mode of operation. This results in reduced cost, complexity and improved security. The classifier is position dependent and can be trained differently for each applicable position in the microgrid. The authors trained several classifiers with different features using a big data set that contained contingencies, and then compared the accuracy of each classifier.

Reducing Number of Communication Devices In Microgrids Using Machine Learning Conference Paper