False Data Injection Attack Detection Based on Local Linear Embedding and Extreme Learning Machine

False Data Injection (FDI) is a new attack method for power system state estimation, which can effectively bypass system monitoring and defense. Due to the highly complex topology of the grid system, the processing of historical data by machine learning method faces the problem of "curse of dimensionality". This paper investigates a new data-driven attack detection method in smart grid based on extreme learning machine (ELM) and local linear embedding (LLE). The LLE automatically extracts the deep features of nonlinear structure in high dimensional data, while ELM is adopted as a classifier to classify the extracted features and detect the attack. The performance of the proposed method is demonstrated through simulation implemented on an IEEE 57-Bus power system. The simulation results validate that the proposed method is able to achieve high accuracy in detecting abnormal data, and also has better generalization and real-time ability compared to other existing learning-based FDI attack detection algorithms.