An Efficient Communication Intrusion Detection Scheme in AMI Combining Feature Dimensionality Reduction and Improved LSTM

Algorithm 2

Using the SAE to implement feature dimensionality reduction algorithm.

	Input: Preprocessed_train, Preprocessed_test, Train_label, Test_label
Output: Encoded_train, Encoded_test, Train_label, Test_label
Load processed data Preprocessed_train, Preprocessed_test
While not reach terminating condition: n-layer autoencoder training (n = 1, 2, 3, 4)
for Epoch in range (1, 100):
/ complete filepath stitching /
os.path.join()
/ Save the model results after each epoch to filepath /
AE_n_point = ModelCheckpoint (filepath, monitor = “val_loss”, verbose = 1, save_best_only = True, mode = “min”)
/ Save the best model to prevent overfitting /
AE_n_stops = EarlyStopping (monitor = “val_loss”, patience = 10, mode = “min”)
break
AutoEncoder_n.load_weights()
Output the prediction result of this layer: layer_n_output, test_n_out
End While
Encoded_train = SAE_encoder.predict (train)
Encoded_test = SAE_encoder.predict (test)
/ save SAE final result /
np.save (Encoded_train, Encoded_test, Train_label, Test_label)
End