A high-speed rail turnout intelligent fault detection method based on hybrid deep learning

Abstract

The invention provides a high-speed rail turnout intelligent fault detection method based on hybrid deep learning. According to the method, a hybrid deep learning method combining deep learning automatic feature extraction and traditional machine learning abnormal point detection is applied to turnout fault detection method research; deep learning is used for automatic feature extraction, featuredata which are smaller in dimension and more abstract are formed, and the problems that the feature extraction process depends on human experience and a clustering algorithm is difficult to calculateunder high-dimensional data are solved; then a clustering algorithm is combined with expert knowledge to select a normal data cluster, and the problem that a large amount of labeled data cannot be obtained is solved; and finally, the data labeled to be normal is used to train a list classification support vector machine for the abnormal point detection, and the problem that no label exists or thelabel is insufficient in the turnout fault detection process is solved.

Description

technical field [0001] The invention belongs to the field of turnout fault detection, in particular to an intelligent fault detection method for high-speed rail turnouts based on hybrid deep learning. Background technique [0002] In recent years, China's high-speed railway has developed rapidly, and now it has the world's largest and highest operating speed railway network, which brings great convenience to people's travel, and its safety has also attracted widespread attention. Turnouts are used to realize high-speed train transfer or cross-line operation, and are key ground signal equipment to ensure the safe operation of railways; their operation status shows the characteristics of large number, frequent operation, and harsh environment, which may easily cause turnout failures and cause train failures. Operational safety hazard. At present, the detection of turnout faults relies on manual judgment. In order to avoid faults, excessive protection and maintenance measures ...

AI Technical Summary

Problems solved by technology

However, in practice, due to the difficulty of manually labeling samples, it is difficult to obtain a large number of labeled samples during the fault detection process of turnouts; there are far more normal samples of turnouts than faulty samples, and the neural network is prone to over-fitting problems during training.

The feature extraction used in the fault detection method of the support vector machine mostly starts from the perspective of geometric parameters. The feature extraction relies on manual experience, and the training process requires a large amount of labeled data, which does not meet the existing reality that there is no accurate fault label for a large number of switch action current curve data. Happening

Images