Reservoir Computing as a Promising Approach for False Data Injection Attack Detection in Smart Grids

Abstract

Smart grids are highly digitalized electricity networks that are increasingly at risk from False Data Injection Attacks (FDIAs), which threaten grid stability. Although traditional machine learning techniques are well‑established for detecting such anomalies, a neuromorphic computing approach like Reservoir Computing (RC) offers a promising alternative. Therefore, this study aims to explore RC for FDIA detection in Smart grids by conducting a Systematic Mapping Study to identify the current research trends and a Benchmark Evaluation of seven models across 21 simulated attack scenarios. The evaluation included the following three metrics: accuracy, robustness, and training‑time efficiency. Findings show that the two traditional approaches included in this evaluation lead with up to 99 % accuracy and minimal training time. Among the five RC approaches, a Delayed Feedback Reservoir with Latency Encoding and Multi-Layer Perceptrons readout achieved ~93 % accuracy (albeit with longer training), while pairing that reservoir with a Logistic Regression readout delivered ~86 % accuracy in under 0.02 s. These findings suggest that appropriately coded and read‑out RC models can serve as robust, resource‑efficient solutions for real‑time FDIA detection.