基于深度学习的电力系统暂态功角与暂态电压稳定裕度一体化评估  

Integrated Evaluation of Power System Transient Power Angle and Transient Voltage Stability Margin Based on Deep Learning

作　　者：史法顺 吴俊勇[1] 吴昊衍 李宝琴 季佳伸 王春明 董向明 SHI Fashun;WU Junyong;WU Haoyan;LI Baoqin;JI Jiashen;WANG Chunming;DONG Xiangming(School of Electrical Engineering,Beijing Jiaotong University,Haidian District,Beijing 100044,China;School of Computer and Information Technology,Beijing Jiaotong University,Haidian District,Beijing 100044,China;Central China Branch of State Grid Corporation of China,Wuhan 430077,Hubei Province,China)

机构地区：[1]北京交通大学电气工程学院,北京市海淀区100044 [2]北京交通大学计算机与信息技术学院,北京市海淀区100044 [3]国家电网公司华中分部,湖北省武汉市430077

出　　处：《电网技术》2023年第2期731-740,共10页Power System Technology

基　　金：国家重点研发计划项目(2018YFB0904500);国家电网有限公司科技项目(SGLNDK00KJJS1800236)。

摘　　要：随着面向高比例可再生能源新型电力系统的转型,系统运行特性日趋复杂。暂态功角稳定(transientangle stability,TAS)与暂态电压稳定(transient voltage stability,TVS)问题相互耦合且频发,为系统安全稳定评估带来严峻挑战。研究首先采用变步长二分法通过调用PSASP从时间维度上构建了暂态电压与暂态功角的稳定边界。研究了不同故障位置、感应电动机占比、负荷率对稳定边界的影响并依托边界确定主导失稳模式。其次提出一种基于注意力机制与一维卷积神经网络融合的电力系统功角稳定及电压稳定裕度评估的新方法。该方法直接面向测量数据,将节点稳态与暂态运行的电压幅值、有功功率、无功功率数据作为输入特征,节省了数据处理时间。通过一维卷积神经网络构建输入特征与极限切除时间的映射,利用注意力机制进一步提高了模型预测效果。通过新英格兰IEEE39节点系统进行分析验证,结果表明该方法可以实现暂态安全裕度的快速评估且具有较高的预测精度。With the transformation of the new power system for high proportion renewable energy, the system operation characteristics are becoming more and more complex. The coupling of transient angle stability(TAS) and transient voltage stability(TVS) occurs frequently, which brings about severe challenges to the system security and stability assessment. Firstly, the variable size dichotomy is used to construct the stability boundary between the TAS and the TVS by the PSASP. Meanwhile, the effects of the different fault locations, the induction motor proportion and the load rate on the stability boundary are studied, and the dominant instability mode is determined based on the boundary.Secondly, a new method for the power system TAS and TVS margin evaluation based on the attention mechanism and the one-dimensional convolutional neural network(1D-CNN) is proposed. This method takes directly the advantage of the measured data, using the voltage, the active power and the reactive power data of node as the input characteristics, which saves the data processing time. The 1D-CNN is adopted to construct the mapping between the input features and the limit resection time, and with the attention mechanism the prediction effect of the model is further improved. Through the analysis and verification of the New England IEEE39 bus system, the results show that this method realizes the rapid evaluation of the transient safety margin and has a high prediction accuracy.

关 键 词：暂态功角稳定 暂态电压稳定 极限切除时间 一维卷积神经网络 注意力机制 

分 类 号：TM721[电气工程—电力系统及自动化]