Impact of False Data Injection Attacks on Machine Learning-Based Cascading Failure Predictions

Machine Learning (ML) and Artificial Intelligence (AI) are garnering popularity in power system failure analysis for their ability to recognize patterns and predict situations based on historical data. The concern over cascading failures in power systems is growing, particularly with the increasing complexity and stochastic nature of modern power grids. While ML techniques have been employed to predict various aspects of cascading failures, the potential for cyber-attacks to infiltrate ML frameworks and manipulate attributes to conceal cascading failures or distort risk assessments is a pressing issue. This paper aims to investigate the impact of false data injection attacks (FDIA) on an example ML-based prediction model for cascading failures. The objective is to understand how these attacks influence ML performance and how such alterations can provide insights to detect such attacks. As anticipated, our assessment indicates that FDIA can alter baseline accuracies, resulting in erroneous predictions of cascade risk.