Cyberattack Detection in Intelligent Grids Using Non-linear Filtering

TL;DR

This paper presents a low-cost, distributed non-linear filtering algorithm for detecting cyberattacks in smart grids, significantly improving security by identifying false data injection attacks with high accuracy and minimal false alarms.

Contribution

It introduces a novel non-linear filtering method based on Kirchhoff laws that is locally implementable, computationally efficient, and enhances existing power system security against cyber threats.

Findings

Detects cyberattacks with 99.9% accuracy

Requires only local data from adjacent nodes

Has a false alarm rate of 4.6%

Abstract

Electric power grids are evolving towards intellectualization such as Smart Grids or active-adaptive networks. Intelligent power network implies usage of sensors, smart meters, electronic devices and sophisticated communication network. This leads to a strong dependence on information and communication networking that are prone to threats of cyberattacks, which challenges power system reliability and efficiency. Thus, significant attention should be paid to the Smart Grids security. Recently, it has been proven that False Data Injection Attacks (FDIA) could corrupt results of State Estimation (SE) without noticing, therefore, leading to a possible mis-operation of the whole power system. In this paper, we introduce an algorithm for detecting cyberattacks based on non-linear filtering by using cyber-physical information from Kirchhoff laws. The proposed algorithm only needs data from adjacent nodes, therefore can be locally and distributed implemented. Also, it requires very low computational effort so that it can be run online, and it is suitable for implementation in existing or new ad-hoc low-cost devices. The proposed algorithm could be helpful to increase power system awareness against FDIA complementing the current SE implementations. The efficiency of the proposed algorithm has been proved by mathematical simulations and computer modeling in PSCAD software. Our results show that the proposed methodology can detect cyberattacks to the SE in 99.9% of the cases with very little false alarms on the identification of spoiled measurements (4.6%).