A Data-Driven Framework for Online Mitigation of False Data Injection Signals in Networked Control Systems

TL;DR

This paper presents a two-stage, data-driven framework utilizing meta learning and deep neural networks to detect and mitigate false data injection in networked control systems, enhancing their security and operational safety.

Contribution

It introduces a novel meta learning-based model selection approach combined with real-time mitigation for false data injection in NCSs, using time series to image transformations.

Findings

Effective false data mitigation demonstrated in mobile robot formation control simulations.

The framework improves system resilience against malicious data attacks.

Meta learning enhances model selection accuracy for diverse data complexities.

Abstract

This paper introduces a novel two-stage framework for online mitigation of False Data Injection (FDI) signals to improve the resiliency of Networked Control Systems (NCSs) and ensure their safe operation in the presence of malicious activities. The first stage involves meta learning to select a base time series forecasting model within a stacked ensemble learning architecture. This is achieved by converting time series data into scalograms using continuous wavelet transform, which are then split into image frames to generate a scalo-temporal representation of the data and to distinguish between different complexity levels of time series data based on an entropy metric using a convolutional neural network. In the second stage, the selected model mitigates false data injection signals in real-time. The proposed framework's effectiveness is demonstrated through rigorous simulations involving the formation control of differential drive mobile robots. By addressing the security challenges in NCSs, this framework offers a promising approach to maintaining system integrity and ensuring operational safety.