电力系统自然频率特性系数区间预测方法  

Interval Prediction Method for Natural Frequency Characteristic Coefficient of Power System

作　　者：蒙永苹 张明媚 向明旭 杨渝璐 黄俊凯 杨知方 MENG Yongping;ZHANG Mingmei;XIANG Mingxu;YANG Yulu;HUANG Junkai;YANG Zhifang(State Grid Chongqing Electric Power Company,Chongqing 400014,China;State Key Laboratory of Power Transmission Equipment&System Security and New Technology(College of Electrical Engineering,Chongqing University),Chongqing 400044,China)

机构地区：[1]国网重庆市电力公司,重庆市400014 [2]输配电装备及系统安全与新技术国家重点实验室(重庆大学电气工程学院),重庆市400044

出　　处：《电力建设》2021年第9期105-111,共7页Electric Power Construction

基　　金：国网重庆市电力公司科技项目“适应重庆电网运行新特性的AGC精细化智能控制策略研究”(SGCQ0000DKJS2000126)

摘　　要：电力系统自然频率特性系数β是整定区域互联电网自动发电控制(automatic generation control,AGC)策略中频率偏差系数B的重要依据.理想情况下,系数B的整定原则是使其恰好等于β系数,从而使系统AGC调节量能准确跟踪系统功率偏差.然而,电力系统β系数具有非线性与时变性,现有B系数整定方法难以有效追踪其变化.对此,提出了基于深度神经网络(deep neural network,DNN)与Bootstrap的自然频率特性系数区间预测方法.该方法利用DNN强大的非线性特征提取能力建立系统功率扰动、备用容量、机组启停方式与β系数间的映射关系,实现β系数的预测,并结合Bootstrap方法得到β系数预测结果的置信区间,可为系统B系数的整定提供有力支撑.算例仿真结果验证了所提方法的准确性与鲁棒性.The natural frequency characteristic coefficient(β) of the power system is a significant basis for setting the frequency bias coefficient(B) in the automatic generation control(AGC) strategy. Setting B equal to β is the ideal principle of the B coefficient setting, because the AGC power adjustment is able to exactly reflect the power mismatch under this circumstance. However, the β coefficient is nonlinear and time-varying. The existing B coefficient setting methods cannot effectively track the changes of the β coefficient. In this regard, the interval prediction method for the β coefficient based on deep neural network(DNN) and Bootstrap is proposed. With the powerful capability of nonlinear feature extracting, DNN is utilized to establish the mapping relationship among power disturbance, reserve capacity, unit commitment, and the β coefficient. Thus, the prediction of the β coefficient can be achieved. In addition, combined with the Bootstrap method, the confidence interval of the predicted β coefficient is further obtained, which provides great supports for setting the B coefficient. Finally, simulation results verify the effectiveness and robustness of the proposed method.

关 键 词：自然频率特性系数 频率偏差系数 区间预测 深度神经网络(DNN) BOOTSTRAP方法 

分 类 号：TM712[电气工程—电力系统及自动化]