SED-Bench: Benchmark for Simulated Electrical Disturbance

In the literature about the problem of power quality, the authors of many articles use simulated data, since it is hard to find benchmarks with real disturbances. For this reason, some authors work with disturbance emulations using electrical devices (e.g., motors and solenoids), that is more close to real anomalies. Moreover, to make a fair comparison with different approaches, for example in the classification process, could be interesting to have a benchmark which could be used for all researchers; however, this was not available yet. In this context, this article has as a contribution: the developing of a benchmark of simulated electrical disturbances, providing it for the community. It has been named SED-Bench, and it contains nine (9) isolated disturbances (interruption, sag, swell, oscillation, impulse, flicker, notching, dc-level, and harmonics) and eight (8) doubled disturbances, four (4) tripled disturbances, and the pure sine (without disturbance). Every signal is available in different data sets with the fundamental frequency of 50 or 60 Hz and SNR values of 20, 30, 50, 60, 70 dB, or without noise. Finally, the developed benchmark is evaluated using an in vitro experiment that uses the Random Forest algorithm to classify part of the data available on the proposed benchmark. In this scenario, we achieve the maximum of 98.89% of accuracy for disturbances with 70 dB of SNR.