Results

Step 1:

Clustering Neighboring PMU Result:

PMU 1, 2, 3, 5, 20 are found to be neighboring PMU, which means they are electrically connected/ close-by.

Step 2:

SVM Result

The training dataset is generated with random value within certain value(1.1 to 1.3) however the attack value of testing data is within range (0 to 3). The accuracy trend of different falsified measurements is shown below.

Multi Layer Neural Network Result:

•The training dataset is generated with random value within same values (1.1 to 1.3).

•The attack vector is varied from 1.01 to 1.05

•The figures show The NN with 4 Hidden layer can detect the attacks till 1.01 p.u.

•The NN with 3 Hidden Layer fails to distinguish attack and events when the attack is stealth i.e. 1.01 p.u.

•To detect stealth attack, 4 layered NN is preferable.

•Another possible solution can be to train the network with wider range of attack vector, which sometimes may be redundant.