基于GMM聚类和PNN的道岔故障诊断研究

当前道岔故障诊断系统大多采用BP神经网络,但由于BP神经网络结构特点,在训练样本大且诊断系统精度要求比较高时,网络常常会呈现出以下不足:不收敛且容易陷入局部最优、常用的数据挖掘方法如小波分析等对数据的利用度不高、从时域或频域角度分析时不够全面和采用数据降维使用的LLE方法会丢失部分有用数据等.采用GMM聚类方法对兰州车...

Research on Turnout Fault Diagnosis Based on GMM Clustering and PNN

Most of the current fault diagnosis systems for turnouts use BP neural network. However, due to the characteristics of the BP neural network, when the training samples are large and the accuracy requirements of the fault diagnosis system are relatively high, the network often has problems such as non-convergence and easy to fall into local optimum. Commonly data mining methods such as wavelet analysis have low utilization of data. The analysis from the time domain or frequency domain is not comprehensive enough and the LLE method used for data dimensionality reduction will lose some useful data. After using the GMM clustering method to classify the 600 sets of power data collected in the computer monitoring system of Lanzhou Station, comprehensive data of the information are selected based on the results to establish the training set and test set of the probabilistic neural network(PNN). The simulation graph from the test set is compared with BP neural network. The results show that the fault diagnosis method based on GMM clustering and PNN can improve the problems of non-convergence and large error.