Research And Implementation Of Temperature State Prediction Of Railway Wagons Bearings Based On Optimizing RVM

Since the beginning of the development of railway transportation,railway wagons have shouldered the burden of bulk cargo transportation.As a key component,rolling bearing has a significant impact on the safe operation of railway wagons.In recent years,China’s railway wagons have entered a period of large-scale maintenance.In the process of maintenance,there are problems of over-repair,resulting in the reduction of freight transport efficiency and the increase of maintenance costs.With the dispatching density of freight train units increasing,it is urgent to realize the transition from "planned repair" to "state repair",that is state prediction of key components,so as to improve the operation and maintenance efficiency of railway freight cars and reduce maintenance costs effectively.At present,temperature change prediction models of rolling bearings on railway wagons have some shortcomings,such as low prediction accuracy.Based on the data accumulated during the running process of rolling bearings,this paper selects the Relevant Vector Machine prediction algorithm based on the non-linearity and high dispersion of the rolling bearing temperature data of railway wagons.(1)Data processing methods such as data cleaning,fusion,and dimensionality reduction are used to preprocess the data,which effectively solve the following problems in the collected data:some key data missed,existing invalid data,multiple different data sources and large data dimension,laying a data foundation for the selection of the prediction algorithm and the construction of the model.(2)Propose the HRVM algorithm.In this paper,the hybrid kernel function is used to improve the accuracy of the prediction results by utilizing the advantages of the obtained kernel function with multiple function characteristics.According to the data characteristics of the rolling bearing temperature data of railway wagons,the Laplace function and the quadratic polynomial are selected to be mixed with a certain weight to form a hybrid kernel function.The kernel function in the traditional algorithm is replaced by the hybrid kernel function,and the HRVM algorithm is proposed.Finally,a certain amount of real data is selected to verify the model.Experiments show that the prediction accuracy of HRVM algorithm is higher than that of traditional RVM algorithm.(3)The ICPSO algorithm is used to optimize the hyper-parameters of HRVM algorithm,and the ICPSO-HRVM algorithm with higher prediction accuracy is proposed.The algorithm solves the problem of randomness and premature maturity in the process of hyper-parameter initialization of HRVM algorithm,and then improves the prediction accuracy of the model.The chaotic particle swarm optimization(CPSO)algorithm does not fully solve the problem of falling into the local optimal solution and mapping unevenly.Therefore,the chaos idea is extended and the chord function is used as the mapping function forming the ICPSO algorithm.The comparative experiment of ICPSO algorithm is carried out,and the experimental result shows that the premature phenomenon of the parameter optimization algorithm is improved.Based on ICPSO-HRVM algorithm,a visual rolling bearing temperature state prediction model is constructed and compared with the traditional algorithm.The results reflect that the prediction accuracy of this-algorithm is higher than that of the traditional algorithm.