Crucial Power Transfer Interface Identification Based on Deep Learning and Bagging Strategy

With the promotion of renewable energy, it has become a more challenging task to keep power systems in secure and stable conditions. In real applications, power transfer interfaces are usually used by operators to monitor and control power systems more efficiently. Therefore, this paper proposes a data-driven approach to identify real-time crucial power transfer interfaces whose security margins are relatively low. First, a deep learning framework is designed as deep features perform better in classification tasks. Then, a new loss function is developed to enhance the data-driven model's generalization ability. Furthermore, bagging strategy is adopted to balance the effect of the missing alarm and the false alarm. Finally, the proposed model is evaluated by the Guangdong Power Grid in China and the simulation results demonstrate that: (1) three corresponding improvements above have been proved, compared with existing methods; (2) the proposed model can identify crucial power transfer interfaces quickly and accurately, and reach the requirements for online applications.