Distributed network malicious attack detection system and method based on convolutional neural network

Abstract

The invention relates to a method for detecting and positioning malicious attack nodes in a distributed network based on a convolutional neural network. The invention specially discloses a distributednetwork internal malicious attack detection and positioning system based on a deep convolutional neural network (DCNN). According to the system, based on a distributed Gossip consensus protocol, a distributed detection and positioning task under a malicious attack of the protocol is set, and an internal attack detection and malicious node positioning decision-making device based on the DCNN is established. According to the detection and positioning method provided by the invention, the characteristics of each node in the distributed communication network can be fully mined, the fitting of a complex nonlinear function is completed, and the accuracy of internal data injection attack detection and malicious node positioning is greatly improved.

Description

technical field [0001] The invention relates to a method for detecting and locating malicious attack nodes in a distributed network based on a convolutional neural network. Background technique [0002] In distributed communication networks, distributed consensus algorithms are very suitable for decentralized multi-task collaborative optimization problems. The algorithm has good robustness and fault tolerance in interactive computing, especially has the ability to accommodate node failures, and can complete collaborative optimization tasks in the case of a small number of node failures and slight noise. However, with the information explosion in the era of big data and the increase in network complexity, consensus-based distributed computing platforms are vulnerable to data injection attacks from internal nodes in the communication network. [0003] The status update behavior of a distributed network is defined as: each node only exchanges data status with its own neighbor ...

AI Technical Summary

Problems solved by technology

[0005] The main disadvantages of traditional methods: (1) Traditional statistical methods have large limitations for nonlinear classification problems, and cannot extract important features hidden in data; (2) The design and demonstration process of detection and localization functions are more complicated and difficult. It is difficult to design a detection and location classifier that meets the needs of the scenario; (3) In the internal attack scenario, the accuracy of detection and location is low, and with the increase of cooperative malicious nodes, the accuracy of detection and location drops more (4) The robustness of traditional statistical methods is poor, and the threshold setting for attack detection and location is relatively complicated

Images