Training datasets are derived from real PMU measurements of the neighboring 5 PMUs.
Each event dataset has 540 samples. In total, we extracts the measurements from 100 events.
For each event dataset, the preprocessing is shown.
Voltage magnitudes are taken as Per Unit (P.U.) which is done by dividing the voltage magnitudes with base value: 550 kV. So that the magnitudes are normalized around 1 p.u.
Attack dataset: 10 event samples are randomly selected and are injected with an attack vector with a range from 1.1V to 1.31V. Those attack randomly targets one of the 5 PMUs.
For each PMU, there are 4 DTWS calculated w.r.t. four other PMUs. The maximum of the four is assigned to that specific PMU. Repeating same process for other PMU for that single event data. We get DTWs for each of five PMUs.
Total 100 such event data have been used. So the final training dataset are 100 x 5 Dataset.
After preprocessing: each attribute in the training data indicates the max(DTW) of one PMU. There is one row per event/attack and one column per PMU. The class label for Event and attack is 0/1.