基于PSO-SVR的同步发电机励磁电流预测  

PSO-SVR based field current prediction for synchronous generator

作　　者：李红连[1,2] 唐炬[2] 

机构地区：[1]成都大学电子信息工程学院,四川成都610106 [2]重庆大学输配电装备及系统安全与新技术国家重点实验室,重庆400044

出　　处：《西北农林科技大学学报（自然科学版）》2013年第6期188-194,共7页Journal of Northwest A&F University(Natural Science Edition)

基　　金：国家重点基础研究发展计划(“973”计划)项目(2009CB724506);国家自然科学基金项目(11205022);国家社会科学基金项目(10XGL0013);重庆市自然科学基金项目(2008BB0327);四川省科技支撑计划项目(2012GZX0083);四川省教育厅科技项目(12ZB170)

摘　　要：【目的】采用粒子群优化支持向量回归(PSO-SVR)模型对同步发电机励磁电流进行预测,为更准确地实现同步发电机转子绕组匝间短路故障的在线诊断提供依据。【方法】以微型同步发电机动模试验的20组正常运行数据作为训练样本,用剩下的13组正常运行数据和33组故障运行数据为检验样本,选取机端电压、有功功率、无功功率为输入量,励磁电流为输出量,通过粒子群优化(PSO)支持向量回归(SVR)的结构和参数,建立PSO-SVR预测模型,进而进行励磁电流预测,并与在线实测的励磁电流进行比较,以误差超过阈值诊断为发生匝间短路故障。【结果】PSO-SVR预测模型的预测误差较误差反向传播(BP)神经网络预测模型小;PSO-SVR模型能设置阈值准确诊断运行状态,而BP神经网络预测模型却不能,并且至少有1次误诊情况出现。【结论】PSO-SVR预测模型的精度优于BP神经网络预测模型,能准确地进行转子绕组匝间短路故障诊断,为同步发电机励磁电流预测、转子绕组匝间短路故障的在线诊断提供了一种新途径。【Objective】 The PSO-SVR prediction method was utilized to predict field current.【Method】 20 sets of normal operation data from micro-synchronous generator dynamic simulation were chosen as training samples,and the left 13 sets of normal operation data and 33 sets of faulty operation data from the simulation were chosen as test sample.Terminal voltage,active power and reactive power were selected as input variables,and field current was selected as input variable.Structure and parameters of support vector regression were optimized by particle swarm optimization algorithm,then the PSO-SVR prediction method was established and field current prediction was conducted.By comparing the predicted field current with the corresponding online measured field current,a rotor winding inter-turn short-circuit fault would be diagnosed if the prediction error（relative error） absolute value exceeded a specific threshold.【Result】 The error of PSO-SVR prediction model was less than BP neural network prediction model,and could set a threshold to diagnose operation condition for a synchronous generator,while the BP prediction model could not set a threshold and error diagnosis happened at least once.【Conclusion】 The PSO-SVR prediction model was better than the BP prediction model,and could accurately diagnose faulty operation condition from normal operation condition for a synchronous generator rotor winding.Overall,the PSO-SVR prediction model was a feasible and effective way for field current prediction and online diagnosis of synchronous generator rotor winding inter-turn short-circuit fault.

关 键 词：支持向量回归 粒子群优化算法 励磁电流预测 同步发电机 转子绕组 匝间短路故障 

分 类 号：TM307.1[电气工程—电机]