Wide-Area damping control enhanced by non-invasive data verification against false data injection attacks

The increasing interconnectivity of power systems and the integration of renewable energy sources into the grid have exacerbated the low-frequency oscillation problem, which directly impacts the stability of the power grid. The performance of wide-area damping control (WADC), which is an effective oscillation suppression technique, is susceptible to false data injection attacks (FDIA). In this study, a non-intrusive data verification method based on an adaptive variational autoencoder (AVAE) and a deep convolutional neural network (DCNN) is proposed to enhance the immunity of WADC to FDIA. The method constitutes a detection module by combining the data dimensionality reduction and reconstruction capability of AVAE and the feature extraction advantage of DCNN. By dynamically adapting to the cyclical changes of grid data through adaptive thresholds, it accurately identifies the behavior of the system in steady and transient states, effectively distinguishes normal and abnormal measurements, and significantly improves the accuracy of performance discrimination and attack detection. In addition, the study proposes an intelligent repair process based on latent representations for anomalous data in the repair module. The module utilizes hierarchical clustering and Gaussian mixture model (GMM) after Bayesian inference to repair the potential space of anomalous data. Then, it accurately repairs the anomalous data to mitigate the impact of attacks through AVAE reconstruction. Simulation experiments on IEEE 39-Bus, 68-Bus, and 118-Bus systems verify that the method can detect network attacks with high accuracy under different attack strengths and noise environments. It can restore manipulated signals with small reconstruction errors to ensure the stability and reliability of the power grid.