HYBRID-CNN: An Efficient Scheme for Abnormal Flow Detection in the SDN-Based Smart Grid

Abstract

Software-Defined Network (SDN) can improve the performance of the power communication network and better meet the control demand of the Smart Grid for its centralized management. Unfortunately, the SDN controller is vulnerable to many potential network attacks. The accurate detection of abnormal flow is especially important for the security and reliability of the Smart Grid. Prior works were designed based on traditional machine learning methods, such as Support Vector Machine and Naive Bayes. They are simple and shallow feature learning, with low accuracy for large and high-dimensional network flow. Recently, there have been several related works designed based on Long Short-Term Memory (LSTM), and they show excellent ability on network flow analysis. However, these methods cannot get the deep features from network flow, resulting in low accuracy. To address the above problems, we propose a Hybrid Convolutional Neural Network (HYBRID-CNN) method. Specifically, the HYBRID-CNN utilizes a Deep Neural Network (DNN) to effectively memorize global features by one-dimensional (1D) data and utilizes a CNN to generalize local features by two-dimensional (2D) data. Finally, the proposed method is evaluated by experiments on the datasets of UNSW_NB15 and KDDCup 99. The experimental results show that the HYBRID-CNN significantly outperforms existing methods in terms of accuracy and False Positive Rate (FPR), which successfully demonstrates that it can effectively detect abnormal flow in the SDN-based Smart Grid.