Smart Meter Data Attacks Assessment: Evaluating Across Diverse Scenarios

Smart meters play a crucial role in transferring information between customers and utility companies by utilizing bidirectional communication and introducing vulnerabilities to cyber-attacks. These cyber-attacks, such as data replay attacks (DRA) and false data injection into databases, introduce errors in power demand load forecasting, which further results in energy supply scheduling problems. This paper provides a learning framework for smart meter cyber attacks by implementing a supervised machine learning algorithm to detect any kind of data anomalies or manipulation. The machine learning model employed in this paper includes Decision Tree, Random Forest, AdaBoost, and XGBoost. The fraudulent dataset is generated by employing two types of attack scenarios, and the performance of each machine-learning model is evaluated across diverse attack scenarios. In this proposed work, the percentage of attack level have been employed as 50%, 30%, 15%, and 5%. The performance metrics employed in this study to analyze the performance of the machine learning model are recall, precision, accuracy, f1-score, and area under the curve. The simulation results show that the attack level or imbalanced data proportion has a significant impact on the performance metrics of AdaBoost as the recall score has decreased from 0.9940 to 0.2608 if the attack level is decreased from 50% to 5%.