Detecting False Data Injection Attacks using NARX-Based Observers in Distributed Cooperative Control for DC Microgrid

In this study, we utilize a Nonlinear Auto-Regressive Exogenous (NARX) observer to estimate voltage and current states within an islanded DC microgrid across normal, load change, and attack scenarios. We implement a distributed cooperative control mechanism to synchronize distributed generation (DG) units for proportional load distribution. Specifically, we introduce a false data injection (FDI) attack to compromise control functions and system stability by intercepting sensor data or communication flows. Detecting such attacks is imperative for system resilience. Additionally, we compare Support Vector Machine (SVM), Neural Network (NN), and NARX methods to assess their estimation capabilities, finding that NARX outperforms in both estimation and attack detection. This study involves the initial operation of the DC microgrid under normal conditions, generating a dataset for training neural networks. This dataset incorporates load variations, enabling the distinction between load changes and potential cyber-attacks. Trained networks are subsequently applied in real-time to estimate the voltages and currents of Distributed Energy Resources (DER) units, allowing for the detection of cyber-attacks based on estimation errors. The efficacy of our approach is verified through MATLAB simulations and real-time validation using the OPAL-RT platform.