基于深度残差收缩网络的电力系统暂态频率安全集成评估  

Integrated Assessment of Power System Transient Frequency Security Based on Deep Residual Shrinkage Network

作　　者：王彦博 吴俊勇[1] 季佳伸 李栌苏 李宝琴 WANG Yanbo;WU Junyong;JI Jiashen;LI Lusu;LI Baoqin(School of Electrical Engineering,Beijing Jiaotong University,Haidian District,Beijing 100044,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

出　　处：《电网技术》2023年第2期482-492,共11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：在我国能源结构加速转型、力争实现“双碳”目标的背景下,传统电力系统也将迎来结构性的转变。其中由于可再生能源的随机性、不确定性和低惯量等特性,大规模新能源并网带来的一系列影响使得电力系统的频率安全问题日益突出。而传统的时域仿真方法在进行评估时有运算量大、计算时间长等缺点,故难以实现实际电力系统灵活多变的运行方式和大量量测数据下的快速评估。为实现对系统频率安全的快速评估,提出一种基于深度残差收缩网络(deep residual shrinkage network,DRSN)的电力系统暂态频率安全集成评估方法。深度残差收缩网络在深度残差网络的基础上引入注意力机制,能够增强有用信息并抑制冗余信息。在此基础上,将样本按最大频率变化率进行划分,并分别采用DRSN网络进行训练构建集成模型。通过引入风电的新英格兰39节点和118节点系统上的仿真结果,表明所用方法与传统深度学习方法相比精度更高,并有着优异的泛化性、鲁棒性和适用性。Under the background of accelerating transformation of China’s energy structure and striving to achieve the goal of "dual carbon", the traditional power system will also usher in structural transformation. Due to the randomness, uncertainty and low inertia of renewable energy, a series of influences brought by large-scale new energy grid make the frequency safety problem of power system increasingly prominent. However, the traditional time domain simulation method has some disadvantages such as large amount of computation and long calculation time, so it is difficult to realize the rapid evaluation of the actual power system under the flexible and changeable operation mode and a large amount of measured data. In order to quickly assess the frequency security of power systems, a power system transient frequency security integrated assessment method based on deep residual shrinkage network(DRSN) is proposed. Deep residual shrinkage network introduces attention mechanism based on deep residual network, which can enhance useful information and suppress redundant information. On this basis, the samples were divided according to the maximum frequency change rate, and DRSN network was used for training respectively. Simulation results on the IEEE 39-bus system and IEEE 118-bus system added with wind power machines show that the proposed method has higher accuracy and excellent generalization、robustness and applicability.

关 键 词：深度学习 电力系统 频率安全 最大频率变化率 深度残差收缩网络 注意力机制 

分 类 号：TM721[电气工程—电力系统及自动化]