FDI Attack Detection in Industrial Control Systems Using 1D-CNN: A Comparative Study on Swat, WADI, And MATPOWER Datasets

Abstract

False Data Injection (FDI) attacks pose a significant threat to the reliability and safety of Industrial Control Systems (ICS), particularly in critical infrastructure such as water treatment and power distribution. This paper proposes a unified machine learning approach using a 1D Convolutional Neural Network (1D-CNN) for detecting FDI attacks across three heterogeneous ICS datasets: SWaT, WADI, and MATPOWER. The architecture is designed to process time-series sensor data through an edge-cloud integrated pipeline, supporting real-time anomaly detection with low latency. Standardized attack simulation and preprocessing ensure consistency across datasets. Experimental results show that the model achieves high accuracy (up to 98.2%), with strong recall and low false positive rates. Moreover, it demonstrates robustness in Remaining Useful Life (RUL) prediction and adaptability to diverse signal environments. The findings validate the effectiveness of a single, lightweight detection model for multi-domain ICS protection. Future directions include deployment via federated learning and integration with digital twin frameworks.