【作者】 徐显明；

【导师】 张江滨；

【作者基本信息】 西安理工大学 ， 水利水电工程， 2007， 硕士

【摘要】 励磁系统是同步发电机的重要组成部分，励磁系统一旦发生故障，直接影响着同步发电机的安全可靠运行。由于发电机励磁功率单元的工作电流很大，作为整流回路中的晶闸管等元件易遭受到过流过压而损坏。因此，快速准确地诊断出功率单元的故障部位和性质，对于缩短发电机的停机与检修时间有着十分重要的意义。论文阐述了发电机励磁系统及其故障检修维护的发展现状，结合当前最新的故障诊断方法，确定了粗糙集-神经网络(RSNN)相结合的故障诊断方法，对同步发电机励磁系统功率单元进行故障诊断。首先，采用MATLAB对励磁系统功率单元的故障进行建模仿真，得出晶闸管发生开路与短路情况下整流回路的输出波形；其次，应用BP神经网络方法对励磁系统功率单元进行故障诊断，并详细介绍了粗糙集理论特点及其应用成果；最后，采用了RSNN方法对励磁系统功率单元进行故障诊断。在RSNN分步法中，先利用粗糙集理论对故障属性建立决策表，并利用粗糙集理论中的知识约简方法化简决策表，得到故障类型的诊断规则并实现功率单元故障类型的诊断，再利用神经网络的非线性映射特性，将训练好的神经网络实现功率单元的故障元诊断；在RSNN整体法中，先对故障电压波形进行数据采样获得神经网络的训练样本，利用粗糙集理论中知识约简，消除冗余的故障样本，再利用约简后的故障样本对神经网络进行训练，将训练好的神经网络用于励磁系统功率单元的故障诊断。通过仿真实验验证，基于粗糙集-神经网络分步法和整体法均能准确的诊断出功率单元的故障元。特别是在RSNN整体法中，经过粗糙集理论处理过的样本数据，使得神经网络的训练规模比单纯的神经网络规模大大减小，在保证故障诊断的正确率不变的情况下，使故障诊断的速度大大提高。更多还原

【Abstract】 The excited system is an important constitute unit for synchronous generator, which would have a direct effect on safety and reliability of the generator, once the excited system goes wrong. Because work current of the power unit of generator excited system is very big, the silicon controlled units of the rectifying circuit are easy to damage by over-voltage or over-current. Therefore, it is very important to diagnose the location and the characteristic rapidly and accurately for shortening stop time and repair time.In this thesis, the development of the generator excited system and the actuality of fault maintenance are summarized. Based on the up to date method of fault diagnosis currently, a fault diagnosis method based on rough-set neural network (RSNN) are bring forward. Firstly, the power unit of generator excited system is modeled and simulated with MATLAB, and educe the export wave shape when the silicon controlled unit is disconnect and direct short; Secondly, the fault of the power unit of generator excited system is diagnosed with BP artificial neural network, and the rough set and it’s experiences that have achieved are introduced detailedly; Finally, RSNN method is adopted to diagnose the fault of the power unit of generator excited system. For the method of RSNN integration by step, first, based on rough set principle,a decision table is established, second, reduce the decision table, then the diagnosis rule of fault type is obtained and the fault type diagnosis of the power unit can be carried, last, according to the characteristic of nonlinear mapping of neural network and the trained neural network, the diagnosis of where the faults are can be carried. For the method of RSNN integration, data of the faulty voltage waveform is collected as a simple for training neural network, then, reduce the simple data, delete the redundancy faulty simple, and train the neural network whit the reduced simple data, finally, apply the trained neural network to fault diagnosis of power unit of the excited system.By simulink test verification, the two methods base on RSNN integration by step and RSNN integration all can diagnose where the faults of power units are, especially, for the RSNN integration, the training size of neural network with rough set become smaller for the sample data compared with single neural network. As a result, not only the accuracy rate of fault diagnosis is ensured, but also the speed of fault diagnosis is enhanced greatly.更多还原