Real-time Detection of False Data Injection Attacks Based on Load Forecasting in Smart Grid

Abstract

Application of computing and communications intelligence has increase the openness and complexity of smart grid to a higher degree. However, this shift also makes smart grids more vulnerable to cyber-attacks. Recently, a new type of invisible attack called false data injection attack (FDIA) has been proposed, which can bypass the existing bad data detection and inject false data into the grid measurements. However, most existing work ignore the potential different purposes of FDIA attacks, and simply assuming the purpose as power theft. In this paper, we model two FDIA attacks based on different purposes, one for economic interests, another for destruction. In order to detect these two attacks, we propose a load forecasting based technique for real-time FDIA detection. Firstly, a support vector regression (SVR) is exploit to forecast the load. According to the predicted results and the system model of the power grid, the measurements of the entire smart grid can be calculate by power flow algorithm. Compared with forecasting measurements directly, the computation cost of this method is very small. Then we train a support vector machine (SVM) to detect the potential FDIA attacks based on deviation between the deduced measurements and the true value. Besides, the injection process is also considered in training phase, thus FDIA attacks can be captured in advance. The performance of our proposed detection mechanism is illustrated through the simulation by IEEE 57-bus test system.