主成分分析与遗传神经网络在制冷系统故障诊断中的应用

Application of Principal Component Analysis and Genetic Neural Network in Fault Diagnosis of Refrigeration System

摘要:

针对低温试验系统制冷设备测点多、数据间存在强相关性等特点,将主成分分析法和遗传神经网络智能识别方法进行组合,引入制冷系统的故障诊断中;结合专家经验和主成分分析客观地对多传感器信息进行了科学合理的故障特征优选,从而确定了神经网络的输入空间;为了克服神经网络易陷入局部最小的缺陷,利用遗传算法的全局搜索能力,对神经网络的初始权值和阈值进行了优化;运用该方法对制冷系统各故障状态进行识别,结果表明,简洁有效的网络结构不仅缩短了训练时间,而且提高了网络的稳定性和分类精度,为监测系统提供了一种有效的故障诊断方法。

Abstract:

According to the characteristics of data measured from refrigeration equipment in low temperature test system, such as a huge number of points, a strong correlation between the data, genetic neural network combined with principal component analysis (PCA) is introduced into fault diagnosis in the refrigeration system. With the knowledge of expert experience and PCA, the fault feature is extracted from multi sensor information in a scientific and reasonable way, so the input space of the neural network is fixed. The defects of neural network is easy to fall into the minimum in local space, but genetic algorithm(GA) has global search ability, aim at eliminating the defects, GA is used to optimize the initial weights and thresholds of neural network. Using the method into the fault state identification of the refrigeration system, it showed that the simple and effective network structure not only shorten the training time, but also improve the network stability and classification accuracy, so it provides an effective method of fault diagnosis for the monitoring system.