Research On Key Problems Of Attack Data Generation And Migration Learning In Oil Depot Industrial Control System

Currently,many studies on security situational awareness of industrial control systems are focused on the security protection of industrial control systems.Once the system is cracked,hackers may break the system by tampering with the underlying business data of the system.Abnormal detection of business data in industrial control system is an important means of system internal security.However,there are two major problems in anomaly detection of industrial control business data:(1)Attack samples from the operation data of industrial control systems are difficult to obtain in the field,resulting in a lack of attack samples;(2)There are multiple sites in the oil depot.The number of oil tanks in different sites is different,and the number of points is different.Anomaly detection models for one site are difficult to be applied directly to other sites.Current anomaly detection model migration is insufficient.The specific work of this paper is as follows:1.Firstly,the open dataset of industrial control is investigated,and the basic data source,composition and point characteristics of the underlying business dataset of industrial control are studied.It is found that such data is two-dimensional time series data with complex industrial control business logic.After that,collect operational data for the bottom of the reservoir.Analyzing the point and time series characteristics of business data based on its business characteristics as the basis for its subsequent research.2.From the time series characteristics of industrial control underlying business data,it can be divided into three types: oscillation,step and pulse.It is difficult to generate samples with the same method.In this paper,an adaptive sample expansion method for industrial control attack is proposed on the basis of the Generative Adversarial Network.Adam optimization algorithm has been improved for stepped time series data generation.The data generated by this method are much more similar than those of Bi LSTM-CNN GAN,GAN and time GAN.3.To solve the problem of poor migration ability of anomaly detection model in oil depot industrial control system,this paper puts forward MLP-Transformer migration learning algorithm.To solve the problem of data heterogeneity between different fields,Multi-Layer Perception is used to map data to the same space;To learn the long-distance association between sample features,a Transformer model based on Multi-Head Attention is used as the main body of anomaly detection;Model migration through fine tuning of frozen layer to improve training efficiency.The evaluation index of migration effect between different reservoir datasets is much higher than traditional migration learning methods such as ADDA and DAN,and is better than the frozen layer fine-tuning effect of convolution neural network.