Detection Mechanism of FDI attack feature based on Deep Learning

At present, with the continuous optimization of grid standards in power system, the degree of convergence of information and communication technologies has become increasingly closer. However, it has also led to the exposure of important infrastructures in the smart grid, especially to the advanced measurement systems in smart metering communication systems. And the most typical type of attacks is False data injection (FDI). Therefore, consumer-centric communication devices, such as Phaser Measurement Units (PMUs) and smart meters, provide powerful support for data acquisition and transmission. As these devices facilitate the acquisition, transmission and consumption of power system data through the integration of communications, some malicious users initiate attacks against measured data in the meter for their own or other economic intention. This kind of attack behavior, on the one hand, affects the availability of meter data. On the other hand, it seriously undermines the authenticity and confidentiality of the measurement data itself. Therefore, in this paper, we proposed attack recognition mechanism based on Deep Belief Network to extract attack features. Our aim is to study the FDI attack behavior and accurately extract the relevant features of the behavior, and provide an effective criterion for the accurate identification of attack behavior in the smart grid. At the same time, through the optimization of the number of nerve cells and the number of layers in each layer of the deep neural network to ensure the real-time detection, the security defense system of the power grid is further enhanced.