Research on Attack Detection Method of Photovoltaic Grid-connected Interface Device Based on Gradient Boosting Decision Tree. (September 2020)

• Record Type: Journal Article
• Authors: Yao, Hao; He, Shengguo; Chen, Rui; Chen, Yuansheng; Zheng, Yucheng; Ding, Kai; Zhu, Xiaofan; Zhou, Guoqi; Huang, Zhiwei; Xiao, Shengyuan; He, Hongyan; Pan, Junjun; Xi, Wei; Yu, Yang; Ning, Baifeng
• Is Part Of: Journal of physics. Volume 1646(2020)
• Publication Date: 2020-09
• Subjects: Physics -- Congresses; 530.5
• DOI: 10.1088/1742-6596/1646/1/012029
• Languages: English
• ISSNs: 1742-6588

Abstract

Abstract: Photovoltaic grid-connected interface devices are an important class of smart devices in microgrids. The authenticity and reliability of the data they acquire, as well as the safety and stability of operation, are related to the safe and reliable operation of the entire microgrid system. However, in the context of microgrid intelligence and informatization, information / network attacks will become the norm, making network-dependent information interaction methods subject to various security risks. The photovoltaic grid-connected interface device involves an open operating environment and is extremely vulnerable to network attacks. The attack information will occupy the space or resources of the photovoltaic grid-connected interface device, making the photovoltaic grid-connected interface device unable to respond to other important requests or instructions in a timely manner, and in severe cases, will cause the device to be paralyzed and affect the normal system operation. Aiming at the above problems, this paper presents an attack detection method based on the gradient-upgraded decision tree model, and gives a detailed design process of attack detection model of the photovoltaic grid-connected interface device. That is, the important data flow in the photovoltaic grid-connected interface device is used as the input of the gradient-upgraded decision tree model, and then the gradient-upgraded decision tree model detect or classify flows, finally, intercept the data flow with attack behavior and give a warning prompt, and forward data without attack behaviors normally.