| All kinds of attacking both inside and outside of power information network system should be prevented along with its widely used. The information secure problem of power information system is even remarkable and it already turns into one important problem which can influence normal running of producing and operating in power system. The network security of power system owns the characteristic of general computer information security and also high security request is even more need to be considered for the infrastructure of national economy. A deep research and exploration work is made in this paper by combining the existing power information network intrusion detection technology based on the analysis of electric power information network invasion threat. The main research work and achievements are as follows:1. Through the analysis of electric power information network structure, in this paper I propose a power information network intrusion detection model based on the deep learning and cloud computing. The model combines misuse detection and anomaly detection methods, which can solve the problem of single misuse detection model can’t detect new attack variants. At the same time to capture the massive power information network data flow, deep learning method for extracting the deep characteristics of the data stream, and in the cloud platform for intrusion detection quickly and accurately.2. By analyzing the various intrusion behaviors of power information network, a new method based on traffic analysis and protocol analysis on Hadoop is proposed. This method can generate the characterization data by analyzing a large number of captured data packets.3. Feature selection is one of the important factors in influencing electric power information network intrusion detection results. This paper proposes a multi-layer auto-encoders feature selection algorithm based on the spark by analyzing the deficiency of the existing feature selection algorithm, combined with the latest research results of machine learning. Through parallel algorithm, which makes power information intrusion detection systems have increased significantly in the speed of feature selection and to reduce the non-response rates.4. Because of the shortcomings of BP network as the classifier in intrusion detection, the BP algorithm based on DBN optimization is proposed. By comparing with other optimization algorithm of BP network, the proposed optimization algorithm can reduce the number of trained BP network and improve classification accuracy in electric power information network intrusion. |
