Enabling Cyberattack-Resilient Load Forecasting through Adversarial Machine Learning

Developing cyberattack-resilient load forecasting is critical for electric utilities in the face of increasingly broad cyberattack surfaces. It is, however, a challenging task due to the adversary's unknown behaviors. This paper bridges the gap by developing an adversarial machine learning (AML) approach for cyberattack-resilient load forecasting. The novelties of this paper include: 1) its analysis of cyber security issues for traditional artificial neural network (ANN) based load forecasting; 2) the ensemble adversarial training it establishes to tackle different attack scenarios; and 3) the selection of parameters for AML it evaluates to achieve desired performance. Test results validate the effectiveness and excellent performance of the presented method.