Statistical outlier detection for diagnosis of cyber attacks in power state estimation

Detection of stealthy attacks on alternating current (AC) static state estimation through false data injection is considered in this paper. To detect the presence of such cyber attacks, we follow a statistical outlier detection approach using a recently proposed machine learning technique called density ratio estimation. The proposed method offers an improved detection especially since our technique does not require attack models when compared to other machine learning techniques implemented in the smart grid literature such as the support vector machine. The simulations are conducted on the IEEE 118-bus system where multiple time scans are merged together to enhance the cyber-security for static state estimation.

I. Introduction

Recently, power systems are evolving towards what is known as the smart grid. The goal consists in integrating more renewable distributed generation (i.e., photovoltaic (PV) and wind) and increasing the energy storage in the grid (for example, via electric vehicles). This will help solve environmental and climatic concerns by preventing the bad effects of released household gas in the atmosphere. The smart grid could be viewed as a cyber-physical system where the cyber part involves communication of data, controls, and diagnoses for the power system in real time. Nevertheless, increasing automatic cyber operations will make the whole power system vulnerable to sophisticated cyber-attacks. Indeed, an intruder could access remotely the grid causing the change in its behaviour. The malicious access could be completed through the Internet and control networks, or by introduced electronic devices such as USB keys to the cyber-system [1].