Fast Contingency Filtering Using Machine Learning for Power System Planning

Modern power systems assess reliability $N -k$ and $k =\{ 2, 3, 4,\ldots\}$ criteria to guarantee the secure, sustainable and optimal operation of power networks. However, performing these studies with traditional methods is computation-intensive and time-prohibitive for the long-term planning of large networks. In this paper, we present a Machine Learning (ML) model for rapid contingency filtering in order to evaluate problematic $N - k$ contingency scenarios. Our proposed model is trained with data sets stochastically created and labeled with AC load flows considering load forecasting. Their inputs are the time and $N - \quad k$ status of network equipment. The performance of the proposed ML model was evaluated on the IEEE 39-Bus System for a planning period of 10yr with a time resolution of 1h. The performance obtained was an accuracy greater than 95% with a time acceleration of approx. 2500x. This result makes the proposed model suitable for supporting decision making during the planning of power systems.