Summary of Model Construction on the Dynamic Detection Accuracy of Railway Lines

This paper for the railway line dynamic detection data analysis, the integrated use of entropy weight method, Topsis evaluation, statistical theorem, RIS rank sum ratio model knowledge, build the quantitative analysis of the reliability of the instrument, the accuracy of the real-time detection data and evaluation interval algorithm model, better solve the problem of railway line dynamic detection data analysis. In view of problem 1, since the line equipment of the dynamic detector is affected by the geometric size of the track, according to the analysis of the relevant literature, the existing data is cleaned and divided into several sub-tables according to the "line-model-vehicle number". In the statistics, the average and median of vertical acceleration and horizontal acceleration are measured by the corresponding dynamic detector under different combination of "line number + model + vehicle number", and four fractions are introduced to reduce the influence of abnormal data. Then calculate the standard deviation of two types of acceleration, using the knowledge of the normal curve, the neighborhood (x, δ), statistics under the range of abnormal data frequency, combined with the entropy weight method and Topsis distance method to build a mathematical model, a comprehensive score, and finally the sensitivity analysis. Finally, 10 unreliable detectors and 3 data errors 1001-101-1230,100-101-101-1356,100102-1580,2001-101-1259,2001-101-1215,2001-102-1502,4001-105-1155. For problem 2, following the problem 1 system, the unreliable detector measurement data will not participate in the data analysis. After trying to make the greedy algorithm combined with linear fitting method and Bayesian formula, the Grubbs criterion was used to build the mathematical model, and then verify the 3-level alarm data after March 1,2024 by using the server to process the data after March 1,2024. In the detection process, the alarm level is above three levels for further analysis. Build a normal data judgment model and write two judgment codes with a test date after March 1,2024 and contain data with test level 3 or above. If either of the two judgment models is normal, the data is normal detection data; otherwise, it is abnormal data. Finally, 55 normal measurement data and 24 abnormal data were obtained. In view of problem 3, the data was processed first, and the algorithm was designed to screen and divide the data. The four lines were divided into the whole line according to the 1 km interval respectively, and the original data set was established. The rank sum ratio model of the whole rank is constructed, and after the variance test and multiple comparisons, the best interval and the worst interval of each line are obtained.