Wind turbine generator gearbox fault early warning method based on fusion model

Abstract

The invention provides a wind turbine generator gearbox fault early warning method based on a fusion model, and the method comprises the steps: extracting the normal operation data of a wind turbine generator in a data collection and monitoring system database, carrying out the preprocessing of the missing and abnormal data through the statistical quartile principle, screening key feature variables through Pearson correlation coefficients, and carrying out the early warning of the fault of the wind turbine generator gearbox. Separately training the extreme gradient boosting tree and the long and short-term memory network model, performing weighted combination and weight calculation on two pieces of prediction time sequence data by adopting an error reciprocal method to obtain a final prediction result, judging the advantages and disadvantages of the model according to error analysis, and finally, performing fault early warning on the gearbox according to the threshold setting of the Mahalanobis distance. According to the method provided by the invention, abnormal data in a big data sample can be effectively removed, key features are extracted, the time for subsequent model training is shortened, the generalization ability is improved through model fusion, an overfitting phenomenon is prevented, and the prediction accuracy of time sequence data is high.

Description

technical field [0001] The invention relates to the technical field of data analysis, in particular to a wind turbine gearbox fault early warning method based on a fusion model. Background technique [0002] When the wind turbine gearbox is in use, the main problems in the field of fault warning are as follows: (1) The operating environment of the wind turbine is affected by the external harsh environment for a long time, and there is a large amount of abnormal noise data in the SCADA data, which needs to be cleared; (2) The wind turbine each There is a coupling relationship between components, resulting in that the prediction dimensions in the SCADA sample data are not completely independent of each other; (3) For the problem of early warning of wind turbine faults, the existing single model often has a low prediction accuracy or a long training time. As a result, it is unable to provide effective early warning information to the operation and maintenance personnel in time....

AI Technical Summary

Problems solved by technology

[0002] When the wind turbine gearbox is in use, there are mainly the following problems in the field of fault warning: (1) The operating environment of the wind turbine has been affected by the external harsh environment for a long time, and there are a large amount of abnormal noise data in the SCADA data, which needs to be cleared; There is a coupling relationship between the components, which leads to the fact that the prediction dimensions in the SCADA sample data are not completely independent of each other; (3) For the problem of early warning of wind turbine failures, the existing single model often has low prediction accuracy or takes a long time to train. As a result, it is impossible to provide effective early warning information for operation and maintenance personnel in a timely manner.

The existing solutions to the above problems are: for the problem of abnormal noise in SCADA data, such as Raida criterion method, density clustering, distance algorithm, etc. can be used to identify and eliminate, but when the sample set has a large amount of data, it often takes a long time; For the coupling problem between multiple characteristic variables, the existing methods include principal component analysis, random forest, neural network, etc., to sort and screen the characteristic variables by key, and eliminate the correlation between the principal components; for the prediction accuracy of the model The problem is not high. In recent years, the method of neural network and deep learning has been used to improve the prediction accuracy, but it takes time to train the model. The method of machine learning can get the prediction result quickly, but the prediction of a single model is accurate. low rate

Images