Voltage-based protection of microgrids using decision tree algorithms

This paper presents a protection method for microgrids by data mining of voltage disturbances. The occurrence of a fault on the system is associated with a sudden voltage depression, which is fast detected by the adaptive cumulative sum (ACUSUM) algorithm. There are other operational (no-fault) events causing voltage depression such as motor starting, transformer energizing, and capacitor or heavy load switching. In order to discriminate between the fault and no-fault events, one cycle of the voltage waveform is preprocessed by the short-time Fourier transform (STFT) to extract and construct effective features of the disturbance. The features are then used in the decision trees (DTs) for the discrimination. The proposed protection method is tested for fault or no-fault conditions of grid-connected or islanded mode of the microgrid operation, as well as radial or meshed topology. The proposed method also identifies the fault type and faulted phase(s) for selective phase tripping. The immunity of the method against different noise levels is investigated. It is shown by the simulation study that by using only two features for symmetrical events and six features for asymmetrical events, any fault can be detected accurately.