Attack Detection and Data Restoration of Remote Estimation Systems Based on D-RCGAN

This paper explores the problem of secure estimation in Cyber-Physical Systems (CPSs) under various network attacks. It proposes an attack detection and data restoration method based on the designed Dual-Constraint Recurrent Conditional Generative Adversarial Network (D-RCGAN). This method utilizes the capability of RCGAN to generate real-valued multidimensional time series by extracting temporal features from conditional data. In addition, to address the issue of data generation accuracy, a dual-constraint approach is proposed to optimize the network parameters of RCGAN. For attack detection, a ratio detector is designed to compare the ratio relationship between generated data and real measurements, enabling the determination of whether the measurements are affected by network attacks. In cases where attacks are detected, the generated data is employed to replace the real data for output, facilitating data restoration. Ultimately, optimal state estimation of CPSs is achieved based on the Kalman filter. Simulation results demonstrate that, compared to other generation models, this algorithm offers better precision in attack detection and efficient data restoration.