基于多尺度特征集的高占比新能源电网连锁故障数据驱动辨识方法  

Data-Driven Cascading Failure Prediction Method for High-Proportion Renewable Energy Systems Based on Multi-scale Topological Features

作　　者：李国庆 张斌 肖桂莲 刘大贵 范慧静 甄钊 任惠 LI Guoqing;ZHANG Bin;XIAO Guilian;LIU Dagui;FAN Huijing;ZHEN Zhao;REN Hui(State Grid Xinjiang Electric Power Co.,Ltd.,Urumqi 830000,China;Department of Electrical Engineering,North China Electric Power University,Baoding 071003,Hebei Province,China)

机构地区：[1]国网新疆电力有限公司,乌鲁木齐市830000 [2]华北电力大学(保定)电力工程系,河北省保定市071003

出　　处：《电力建设》2023年第6期91-100,共10页Electric Power Construction

基　　金：国家自然科学基金青年科学基金项目(52007092,51107040);国网新疆电力公司科技项目(SGXJ0000TKJS2100234)。

摘　　要：随着我国新型电力系统建设进程的不断推进,电力系统运行工况的不确定性大大增加,抗扰动能力更低,连锁故障发生率更高,演化时间更短。传统的基于潮流计算的方法依赖对场景的枚举考虑随机因素,时效性和准确率都难以应对高占比新能源电网复杂的运行工况,因此,提出一种挖掘历史数据规律的基于多尺度特征集的高占比新能源电网连锁故障数据驱动辨识方法,充分考虑历史运行数据中包含的随机信息。从宏观、中观、微观3个层次提取能够表征复杂网络拓扑特征与系统运行状态的特征指标,构成特征指标集;基于双向长短期记忆神经网络学习历史连锁故障过程中复杂网络特征指标与系统运行状态间的映射关系,构建高占比新能源电网连锁故障辨识模型,并通过新疆电网拓扑验证了所提模型的有效性。With the continuous advancement in the renewable power system construction process,the uncertainty of system operating conditions has significantly increased,resulting in a lower antidisturbance ability,higher cascading failure occurrence rate,and shorter evolution time.The traditional power-flow calculation method relies on the enumeration of scenarios to consider random factors,making it difficult to cope with the complex operating conditions experienced with a high percentage of new energy grids in terms of timeliness and accuracy.Therefore,a multi-scale feature-set-based datadriven method for identifying cascading faults under conditions of having a high percentage of new energy grids is proposed by mining historical data patterns and fully considering the stochastic information contained in historical operation data.Indexes that can describe the topological characteristics of a complex network and the operation state of the system are extracted from the macro-,meso-,and micro-levels to form the index set.Based on a bidirectional long-term and short-term memory neural network,the mapping relationship between the index set and the system operation state is studied using the historical cascading failure dataset.This enables a cascading fault prediction model of a high-proportion renewable power system to be constructed.The effectiveness of the proposed model is verified using the topology of the Xinjiang power-grid system.

关 键 词：连锁故障 数据驱动 拓扑特征 高占比新能源 复杂网络 

分 类 号：TM732[电气工程—电力系统及自动化]