Research on Anomaly Detection Model for Power Industrial Control Networks Based on Bidirectional Recurrent Neural Networks

In response to the current challenges of low accuracy in detecting abnormal data traffic and insufficient sample space in power industrial control networks, we propose a novel approach. Our method involves leveraging bidirectional recurrent neural networks(Bi-RNN) with long short-term memory (LSTM) architecture for anomaly traffic detection. Additionally, we introduce a sample augmentation technique based on generative adversarial networks (GAN) to address the limited sample size issue. We train our model on training datasets with temporal characteristics and evaluate its performance on testing datasets, achieving a prediction accuracy exceeding 99.6%. Moreover, through the sample augmentation model, we expand the dataset by over 100,000 samples.