【作者】 赵永玲；

【导师】 任伟建；

【作者基本信息】 大庆石油学院 ， 控制理论与控制工程， 2003， 硕士

【摘要】 随着技术的发展，控制系统越来越复杂，因而及时、准确地诊断出系统的故障，保证其平稳可靠地运行也就变得越来越重要。本文介绍了故障诊断的任务，研究了控制系统中存在的故障，以及它们的数学表示方法。 基于神经网络控制系统的故障诊断分为两部分，一是故障样本数据及检验数据的采集；二是故障诊断。在故障样本数据及检验数据的采集过程中，首先建立了控制系统的数学模型，人为地让控制系统发生各种故障，从而采集到各种故障数据，经过归一化处理后，作为训练神经网络的样本数据；同时，也采集检验数据，用以检验训练出的神经网络是否能够起到故障诊断的作用。在故障诊断过程中，采用BP网络，针对BP算法收敛速度慢、容易陷入局部极小的缺点，采用了“成批处理”的学习方法，这种方法在训练神经网络的过程中，不受学习样本排序的影响，使收敛速度加快；采用了改进的BP算法，采用共轭梯度法，在迭代过程中增加了惯性量；同时采用了学习率自调整的方法，从而使收敛速度和故障诊断精度都有提高。 应用本文所讨论的方法利用采集到的电流数据对抽油机井的故障进行了诊断，实际结果表明，基于神经网络的故障诊断方法可以用于实际生产过程中。更多还原

【Abstract】 With development of technology, control system is more and more complicated. So it becomes more and more importance, diagnosing accurately fault in time and ensuring that system runs reliably. The paper introducing the tasks of fault diagnosis, discuss the fault in the control system and their mathematical representing methods.In the control system of neural network, the fault diagnosis consists two parts: one is the collection of fault sample data, the other is fault diagnosis. So during the collection of fault sample data and checking data, the paper sets up the mathematics model of control system first. We cause all kinds of fault in the control system, then collects a variety of fault data. After unitarily dealing, the fault data server as the sample data of training neural network. At the same time, according to the collection of checking data, it examines whether the trained neural network have the function of fault diagnosis. In the process of diagnosis, as the response of the slow constringency rate and disadvantage that error easily gets into partial minimal value of BP algorithm in using BP network. It adopts three solutions. First, using the learning method of "batch dealing". During the training neural network, this method accelerates the speed of constringency, neglecting the effect of studying sample order. Second, adopting the improving BP arithmetic, Conjugated Gradient Method. It adds the inertia in the process of iterative. Third, using the learning rate adaption, it improves the speed of constringency and accurate rate of diagnosis fault.By using the method of this paper, we collect the data of a pump-jack electric current and diagnose it faults. The result proves, the diagnosis method on fault by neural network can be used in reality.更多还原