Ensemble Voting-Based Anomaly Detection for a Smart Grid Communication Infrastructure

Advanced Metering Infrastructure (AMI) is the metering network of smart grid that enables bidirectional communications between each consumer’s premises and provider’s control center. The massive amount data collected supports real-time decision-making required for diverse applications. communication infrastructure relies on different types, including Internet. This makes vulnerable to various attacks, which could compromise security or have devastating effects. However, traditional machine learning solutions cannot adapt increasing complexity diversity attacks. objective this paper develop an Anomaly Detection System (ADS) based deep using CIC-IDS2017 dataset. dataset highly imbalanced; thus, a two-step sampling technique: random under-sampling Synthetic Minority Oversampling Technique (SMOTE), proposed balance system utilizes multiple hidden layer Auto-encoder (AE) feature extraction dimensional reduction. In addition, ensemble voting both Random Forest (RF) Convolutional Neural Network (CNN) developed classify multiclass attack categories. evaluated compared with six state-of-the-art algorithms: (RF), Light Gradient Boosting Machine (LightGBM), eXtreme (XGboost), (CNN), Long Short-Term Memory (LSTM), LSTM (biLSTM). Experimental results show model enhances detection class other models overall accuracy (98.29%), precision (99%), recall (98%), F1 score UNDetection rate (UND) (8%).