一种基于深度自适应网络迁移的暂稳评估模型更新框架

An updating framework of a model for transient stability assessment based on a deep adaptive network transfer

为解决电力系统的运行方式或拓扑结构变化后暂稳评估模型的适应性问题,常规的特征迁移学习方法主要侧重于拉近源域与目标域数据集间的条件分布或边缘分布的距离,却不能定量的评价这两种分布对于不同域之间的贡献,导致模型迁移性能不理想。针对该问题,引入SENet注意力机制和动态分布自适应算法,构建了基于SEDDAN迁移的深度自适应网络暂稳评估模型更新框架,从特征提取和不同域间分布权重的动态调整两个层面进行改进,进一步提升了评估模型的迁移性能和自适应性。在IEEE 39和IEEE 140节点系统上进行测试,仿真结果表明所提模型在更新后的评估准确性、适应性和迁移性能方面有一定的优势。

To solve the adaptability problem of transient stability assessment models after the changes in the operation or topology of power systems, the conventional feature transfer learning methods mainly focus on bringing the conditional or marginal distributions between the source and target domain datasets closer together, but fail to quantitatively evaluate the contribution of the two distributions to different domains, resulting in unsatisfactory model transfer performance. To address this problem, SENet attention mechanism and dynamic distribution adaptive algorithm are introduced, and a deep adaptive network transient stability assessment model update framework based on SE-DDAN transfer is constructed,which is improved from two aspects, namely, feature extraction and dynamic adjustment of distribution weights between different domains, to further enhance the transfer performance and adaptability of the assessment model. The model is tested on IEEE39-bus and IEEE140-bus systems and the simulation results that the proposed model has advantages in assessment accuracy, adaptability and transfer performance after updating.