Power System Malware Detection Based on Deep Belief Network Classifier

Abstract

In order to achieve accurate detection of unknown malware in power system, this paper proposes a malware detection system based on Deep Belief Network (DBN). The system deconstructs the malware into an opcode sequence, extracts the feature vector with the detection value, and uses the DBN classifier to classify the malicious code. Through the experiments of classification performance, feature extraction and unlabeled data training, it is proved that DBN-based classifiers can use unlabeled data for training and have better accuracy than other classification algorithms. The DBN-based automatic encoder can effectively reduce the dimension of the feature vector significantly.