Safety Risk Monitoring of Cyber-Physical Power Systems Based on Ensemble Learning Algorithm

Abstract

The traditional security risk monitoring technology cannot adapt to cyber-physical power systems (CPPS) concerning evaluation criteria, real-time monitoring, and technical reliability. The aim of this paper is to propose and implement a log analysis architecture for CPPS to detect the log anomalies, which introduces the distributed streaming processing mechanism. The processing mechanism can train the network protocol feature database precisely over the big data platform, which improves the efficiency of the network in terms of log anomaly detection. Moreover, we propose an ensemble prediction algorithm based on time series (EPABT) considering the characteristics of the statistical log analysis to predict abnormal features during the network traffic analysis. We then present a new asymmetric error cost (AEC) evaluation criterion to meet the characteristics of CPPS. The experimental results demonstrate that the EPABT provides an efficient tool for detecting the accuracy and reliability of abnormal situation prediction as compared with the several state-of-the-art algorithms. Meanwhile, the AEC can effectively evaluate the differences in the cost between the high and low prediction results. To the best of our knowledge, these two algorithms provide strong support for the practical application of power industrial network security risk monitoring.