Power System Frequency Safety Assessment Scheme: Multi-Branch Learning Method Based on Ensemble Full Connection

Facing the frequency safety problem caused by the large-scale application of renewable energy in power systems, this paper proposes a frequency safety assessment scheme (FSAS) for power systems based on a novel deep learning structure. Firstly, a novel data preprocessing method is proposed, which takes the feature data blocks of the same attribute as the normalization object, which can effectively improve the performance of the evaluation scheme. Then, referring to the idea of ensemble learning, a multi-branch learning network based on ensemble full connection is designed. This network uses a multi-branch structure to fully mine and learn the deep features from different aspects and uses the ensemble full connection structure to integrate these features and further fit them. Finally, a multi-task FSAS with a parallel structure is proposed, which achieves the dual goals of simultaneously evaluating the Frequency Response Safety Level and Frequency Response Safety Time. Taking the IEEE 39 bus system and the Illinois 200 bus system as examples, both contain renewable energy, the example test proves the effectiveness of the data processing method and the rationality of FSAS structure, and the comparative experiment shows that it has the highest accuracy. Moreover, FSAS has good anti-noise, robustness.