面向稳控策略校核的并联图注意力网络稳定评估模型  

Parallel Graph Attention Network Based Transient Stability Assessment Model for Stability Control Strategy Checking

作　　者：朱思婷 管霖[1] 黄济宇 陈鎏凯 ZHU Siting;GUAN Lin;HUANG Jiyu;CHEN Liukai(School of Electric Power,South China University of Technology,Guangzhou 510641,Guangdong Province,China)

机构地区：[1]华南理工大学电力学院,广东省广州市510641

出　　处：《电网技术》2023年第9期3836-3846,共11页Power System Technology

基　　金：国家自然科学基金项目(52077080)。

摘　　要：考虑稳控策略的数据驱动稳定评估模型可以实现稳控策略的快速评估与校核,是值得探索的研究方向。应用中不仅要求稳定评估模型准确判断系统失稳与否,还需要获得主导失稳机群等信息,辅助稳控策略调整。论文提出了一种并联结构的图深度学习稳定评估模型。分别采用图点积注意力网络和时序卷积网络,有效提取故障期间和稳控动作前后的关键特征。设计了发电机分段稳定指标,并通过机组级稳定指标和机组稳定分类指标的多任务并行预测和相互校核,实现了高精度和细粒度的稳定评估。在IEEE10机39节点系统算例的对比分析验证了模型的有效性和准确性。The data-driven stability assessment model provides rapid evaluation and verification of the stability control strategies,a valuable research direction.In the practical application,it is required that a stability evaluation model can not only estimate the stability state of a power system accurately,but also provide more information about the leading instability generators to assist the adjustment of stability control strategies.In this paper,a graph-deep-learning based stability evaluation model with the parallel structure is proposed.The dot-product graph-attention-network and the temporal convolution network are applied separately to extract the kernel features under the impacts of the fault and that before and after the stability control actions effectively.A piece-wise stability index for the generators is designed.The stability evaluation results with high precision and fine granularity can be obtained through the multi-task parallel prediction and the mutual verification among the units’stability index and the units’stability indices.The effectiveness and precision of the proposed model are verified in the IEEE 39-bus system.

关 键 词：暂态稳定评估 点积注意力 并联结构 图深度学习 

分 类 号：TM721[电气工程—电力系统及自动化]