Malicious code dynamic traffic camouflage detection based on deep reinforcement learning in power system

Abstract

In order to solve the problem that malicious code intrudes into software in various forms, which leads to its security performance degradation and cannot be used normally, this paper proposes a malicious code dynamic traffic camouflage detection method based on deep reinforcement learning in power system. The average mutual information between codes is calculated by deep reinforcement learning, and the weighted information gain of each code type feature is obtained. Different types of code feature set classifiers are generated, and an optimal classifier is output for each type of code feature set. The features are reduced by Linear Discriminant Analysis (LDA), and the network code is classified according to the extracted features. The potential malicious code is detected according to the explicit rules of deep reinforcement learning. Simulation results show that the detection method can improve the accuracy of malicious code classification, and the detection performance is increased to about 35%.