compressor fault diagnosis method based on XGBoost feature extraction

Abstract

The invention discloses a compressor fault diagnosis method based on XGBoost feature extraction. The method comprises the following steps: firstly, according to fault data and a fault type, customizing a loss function of an XGBoost algorithm, and iteratively constructing a fault split tree; secondly, extracting leaf node position index vectors of the samples in the fault tree, carrying out featurecoding reconstruction, and acqiirng intelligent representation of hidden fault information; and then, based on the representation matrix, respectively establishing fault prediction models by using anSVM (Support Vector Machine) algorithm and a neural network algorithm to realize prediction and diagnosis of multiple fault modes. The method has the characteristics that hidden fault feature information in the data can be fully mined, so that the fault diagnosis and prediction precision is higher.

Description

technical field [0001] The invention relates to a fault diagnosis method based on machine learning, in particular to a compressor fault diagnosis method based on XGBoost feature extraction. Background technique [0002] For the fault diagnosis model, the quality of feature extraction largely determines the performance of the model, which is the key link in fault diagnosis. Feature extraction is the process of mining and extracting the most representative information from fault data, and it is also a process of deeply mining the hidden information in fault data. Poor fault characteristics not only affect the efficiency of algorithm operation, but also reduce the prediction accuracy of the algorithm model for faults. Therefore, it is crucial to study effective feature extraction methods. [0003] At present, most literatures divide fault diagnosis methods into analytical model-based methods, empirical knowledge-based methods, and data-driven methods. As the current system h...

AI Technical Summary

Problems solved by technology

[0004] Among them, in fault diagnosis based on machine learning, traditional intelligent learning methods, whether used for classification or regression, are mostly shallow structure algorithms, and their limitations lie in the limited ability to express complex functions in the case of limited samples and computing units. For complex classification problems, its generalization ability is restricted to some extent. How to mine and express fault features from monitoring data is the research difficulty of this type of method. If the hidden information in fault data can be reasonably extracted and represented, it will be Better fault detection and prediction results

The current machine learning methods used in the field of fault diagnosis all fit the monitoring data from the perspective of approximation theory, and there are deficiencies in approximation accuracy. For example, neural networks, support vector machines, etc., are not yet able to fully mine the faults in the monitoring data. feature

Images