Research On High-Speed Train Working Condition Recognition Based On Multi-Scale Entropy Of Frame Load Signal

With the growth of the national economy,the way people travel has changed,from a single car and bicycle to a fast plane and high-speed rail.According to statistics,the number of high-speed rail passengers in my country in 2015 was only 2.5 billion,and the number of high-speed rail passengers in 2019 increased rapidly to 3.5 billion.The continuous increase in the number of passengers fully proves the universality of highspeed rail and also raises serious concerns about the safety of high-speed rail operations.The test.The increase in the operating speed of high-speed railways and the increase in passenger capacity put forward new requirements for the fatigue reliability of rail vehicle components.As a key load-bearing component of EMUs,bogies are critical for structural fatigue strength evaluation and reliability design.The compilation of the load spectrum is the prerequisite for studying the structural reliability of the key components of the EMU bogie.The compilation of large-scale load spectrum requires known high-speed EMU bogie frame load data under different operating conditions.Therefore,by analyzing the signal characteristics of the frame load under different operating conditions,the statistics that can characterize the characteristics of the operating conditions are found.On this basis,the identification of typical working conditions is carried out.This paper selects the actual load signal of the Chinese standard EMU on the Daxi Line for analysis.The main research contents are as follows:(1)In view of the fact that the traditional time-domain signal analysis method cannot accurately describe the small characteristics of the signal that changes in frequency with time,this paper uses wavelet transform and integrated empirical mode decomposition in time-frequency joint analysis to process the original load data in time-frequency.For a single scale that is difficult to comprehensively summarize the time series complexity and the unknown time scale,this paper uses multi-scale entropy to study the load complexity of the frame under different working conditions,and obtain the change law of the load complexity of different working conditions.(2)Based on the complexity law of frame loads under different working conditions,a multi-scale entropy-based feature extraction method for working conditions is established,and the extracted features are eliminated by increasing the amount of data for significance testing.Existing studies have shown that the pattern recognition results do not continue to increase with the increase in dimensionality.Too many feature dimensions will lead to long recognition time and reduced recognition accuracy.Therefore,the Relief algorithm is used to reduce the dimensionality of the extracted working condition feature array.,Establish the feature vector.In order to prove that the working condition feature extraction method based on multi-scale entropy has a positive effect on improving the working condition recognition effect,a working condition feature vector based on traditional time-frequency analysis is established to prepare for the subsequent comparison of the traditional and improved working condition recognition results.(3)The established working condition feature vectors based on traditional timefrequency analysis and multi-scale entropy are respectively used as input vectors and substituted into the support vector machine for pattern recognition.The traditional optimization algorithm grid search and the natural evolution algorithm genetic algorithm are respectively used to optimize the support vector machine kernel parameter g and the penalty coefficient C,and the best is selected according to the recognition result.In the end,the recognition results of braking condition,turnout condition and straight curve condition based on multi-scale entropy are all above 90%,which basically meets the engineering requirements.The recognition results of turnout conditions based on traditional time-frequency analysis are only 50%～60%,which proves that the feature extraction method based on multi-scale entropy can significantly improve the recognition effect of turnout conditions.57 graphs,32 tables,75 references...