基于离散Hopfield网络的船舶发电机故障诊断研究

【作者】 徐若冰；

【导师】 施伟锋；

【作者基本信息】 上海海事大学 ， 控制理论与控制工程， 2007， 硕士

【摘要】 船舶发电机作为船舶电力系统的主要设备,对整个电力系统的安全有效工作起到了至关重要的作用。随着船舶发电机容量的不断增大,船舶营运对发电机的安全和可靠性提出了越来越高的要求。在船舶发电机故障中,定子单相接地故障、机端相间短路故障、失磁故障和定子某相绕组短路故障都是船舶发电机的常见故障。因此对船舶发电机的这些故障进行诊断具有重要的理论意义和工程实用价值。在研究船舶发电机数学机理建模的基础上,在深入剖析Simulink的PowerSystem中发电机模型内部结构的基础上,根据同步发电机典型故障的发生机理,进行建模与仿真。采集相电流有效值作为特征参数,并经过快速傅立叶变换(FFT)等样本处理手段,得到神经网络的学习样本,作为神经网络输入。本文利用离散Hopfield(DHNN)网络作为内容寻址存储(content-addressedmemory device—CAM)表现出来的联想记忆能力,将5种故障状态下的学习样本存储为DHNN的5种模式,再运用DHNN在存储的典型模式中寻找到唯一一个与给定测试样本最相似的模式。从而确定测试样本所代表的故障。软件仿真与测试结果显示,网络可以对几种船舶发电机典型故障进行有效地识别。本文还对离散Hopfield网络的容错能力做了分析,通过对测试样本进行不同程度的噪声污染得到该网络能成功诊断的概率。这样一方面体现了离散Hopfield网络的优越的容错能力,另一方面也为以后对该网络进行进一步优化提供了相关数据。更多还原

【Abstract】 As the main equipment of power system, shipboard power generator plays a crucial role of the security and stability for entire power system. Along with the increase of the capacity of shipboard power generator, people’s demand for safe operation and reliability of generators is greater and greater. Among all the shipboard generator faults, single-phase stator ground fault, the terminal phase short-circuit fault, Magnetic loss fault and single-phase stator windings short-circuit fault are common faults of shipboard generator. So diagnosis for these faults of shipboard generator has important theoretical and practical value.Based on the research of ship generators mathematical mechanism, the in-depth analysis of internal structure of generator model in Simulink and the reason of typical faults of synchronous generator, this paper has done the modeling and simulation. RMS current phase was collected as a sample. After several sample processing means such as Fast Fourier Transform (FFT), this paper got the study samples of neural network for input vectors.By using the associative memory capacity of Discrete Hopfield Neural Network(DHNN) as a content - addressed memory(CAM)device, DHNN saved study samples of 5 faults as 5 typical styles and, found the corresponding one which was most similar to the given test samples in study samples saved before. So that the style of fault that test samples stand for were identified and, the purpose of fault diagnosis was also achieved. Test results show that the network can effectively identify several common faults of shipboard generator.The paper also analyzed the fault-tolerant ability of discrete Hopfield network. Based on polluting sample testing by varied degrees, this paper got the probability of success diagnosis. In this way, this paper can show the superior fault-tolerant ability of Discrete Hopfield Neural Network in the one hand; in the other hand, it can also support the relevant data for further optimization in the future.更多还原