基于堆叠自动编码器的电力系统暂态稳定评估  

Transient Stability Assessment Based on Stacked Autoencoder

作　　者：朱乔木 陈金富[1] 李弘毅 石东源[1] 李银红[1] 段献忠[1] ZHU Qiaomu;CHEN Jinfu;LI Hongyi;SHI Dongyuan;LI Yirthong;DUAN Xianzhong(State Key Laboratory of Advanced Electromagnetic Engineering and Technology （School of Electrical and Electronic Engineering, Huazhong University of Science and Technology）, Wuhan 430074, Hubei Province, China;State Grid Hunan Electric Power Corporation Economic and Technology Research Institute, Chan~sha 410004. Hunan Province. China)

机构地区：[1]强电磁工程与新技术国家重点实验室(华中科技大学电气与电子工程学院),湖北省武汉市430074 [2]国网湖南省电力公司经济技术研究院,湖南省长沙市410004

出　　处：《中国电机工程学报》2018年第10期2937-2946,共10页Proceedings of the CSEE

摘　　要：将深度学习的思想和模型引入电力系统暂态稳定评估研究中，提出一种基于堆叠自动编码器的电力系统暂态稳定评估方法。该方法无需人工计算形成输入特征，直接面向底层量测数据，通过深层架构建立量测数据与稳定类别之间的非线性映射关系。采用一种“预训练一参数微调”的两阶段学习方法，同时引入稀疏化技术和Dropout技术对模型参数进行优化。训练后的模型能够依靠深层结构挖掘数据的隐藏模式，提取出有利于暂态稳定评估的高阶特征。此外，该方法能够通过大量无标注样本的无监督训练提高模型泛化能力，从而大大缩减训练样本时域仿真耗时。新英格兰10机39节点系统上的仿真结果表明所提方法比常规浅层评估方法的评估性能更加优越。This paper introduced the concept and model of deep learning into the field of transient stability assessment （TSA） and presented a novel TSA model based on stacked autoencoder （SAE）. Instead of the hand-crafting features, the underlying measurements during fault duration were employed as the inputs. The stability statuses of power systems were used as the outputs of the SAE model. The parameters of SAE model were optimized by a two-stage training strategy, i.e., pre-training firstly and followed by fine-tuning, and incorporated with the sparse technic and dropout technic. The SAE model is capable of excavating the hidden mode of input data, and automatically extracting the high-order features conducive to TSA. Furthermore, the TSA performance of SAE can be enhanced by unsupervised training employing no-label samples, which significantly reduces the computation burden and time consuming of time-domain simulations. Experiment results on New England 39-bus system demonstrate that the proposed method outperforms common TSA methods.

关 键 词：深度学习 电力系统 暂态稳定评估 堆叠自动编码器 底层量测数据 

分 类 号：TM743[电气工程—电力系统及自动化]