Research On Abnormal Detection And Life Prediction Method Of Power Supply Vehicle

This research aims at the problems of power supply vehicle systems,such as the lack of minor anomaly diagnosis and the inability to accurately obtain the remaining service life.With the help of the 120 KW military power supply vehicle system simulation platform developed by the previous team,combined with those network ways which are used to study the abnormal diagnosis and residual life prediction of the power supply vehicle system,such as Deep Belief Networks(DBN),Transfer Learning,Principal Component Analysis(PCA),Support Vector Regression(SVR),and Gate Recurrent Unit(GRU),etc.The research results of system abnormal diagnosis and remaining life prediction have important application value for improving the health management level of power supply vehicle system and other complex systems.The main work and contributions are as follows:1)Research on abnormality diagnosis method of power supply vehicle based on DBN and transfer learning.In view of the shortage of abnormal state data and the lack of obvious characteristics,this research proposes an abnormal diagnosis method based on transfer learning.Based on the established DBN fault diagnosis model,this method uses transfer learning and less abnormal data to obtain the abnormal diagnosis model of the power supply vehicle through further training.Through the comparative experiment on the power supply vehicle simulation system,the results show that the combination of transfer learning is more direct to use the classic deep learning method,which has a higher accuracy for the abnormal diagnosis in the case of data not sufficient,and to a certain degree provides a guarantee for improving the safety and reliability of the power supply vehicle.2)Research on life prediction method of power supply vehicle based on PCA-SVR.In this research,a PCA-SVR based life prediction method is proposed for the case that the degradation data of the power supply vehicle system is miscellaneous,nonlinear and noisy.This method is based on PCA to reduce dimension and noise of power supply vehicle system degradation data,and SVR to establish the life prediction model of power supply vehicle system,thereby completing the prediction of the remaining life of power supply vehicle system.The experimental results show that the PCA-SVR based life prediction model can obtain more accurate life prediction value under the condition of less training samples,which can provide reference for the health maintenance of power supply vehicles to a certain degree.3)Research on life prediction method of power supply vehicle based on GRU network.In order to further accurately predict the remaining service life of the power supply vehicle system,considering the data in the degradation process of the power supply vehicle system,not only has its own important value,but also has the time series correlation between the data,this research proposes a power supply vehicle life prediction method based on GRU network.Based on GRU network,the time series model of power supply vehicle state is established,which can effectively obtain the important characteristics of data itself and between data.The experimental results show that the service life prediction model based on GRU network can obtain the prediction value which is very close to the actual remaining service life under the condition of sufficient training samples,which provides a more meaningful basis for improving the health maintenance of power supply vehicles.