Detection of False Data Injection Cyber-Attack in Smart Grid by Convolutional Neural Network-Based Deep Learning Technique

Abstract

Electric Power Grid is getting evolved with the aid of advanced sensors, modern communication system, and cutting-edge computation capabilities enabling smart grid operations. At the same time, it is getting more vulnerable to cyberattack such as False Data Injection Attack (FDIA). The critical control process of smart power grid depends on correct State Estimation (SE). A FDIA targets the measurements which are used to calculate SE, thus result in erroneous SE calculation. This results in disrupting the control system, causing havoc to the power transmission. An attacker with some information of the system may perform FDIA which cannot be detected by conventional threshold-based detection system. So, Machine Learning (ML)-based systems are being proposed for detection of FDIA. In this paper, some ML-based models, e.g., Support Vector Machines (SVM), Light Gradient Boosting Machine (LGBM) are prepared, tuned and their performance are tested. Some Deep Learning (DL)-based models, e.g., Convolutional Neural Network (CNN), Artificial Neural Network (ANN) are also experimented with. The CNN models show better results, so a number of configurations are tested on. Finally, a CNN-based model is proposed for detection of FDIA in smart grid in this paper. The models are tested on data related to IEEE 14 BUS. The results of the experiment represent that the proposed Deep Learning-based CNN model can detect FDIA in electric power grid with high accuracy.