Research On Quality Control Algorithm Of Wind Speed Observation Data Along High-speed Railwa

The observation data of wind speed along high-speed railway will be interfered by various factors in the process of collection and transmission,leading to the existence of low-quality data in wind speed data.On the one hand,the high-speed railway wind monitoring system sends out wind alarm information based on the real-time monitoring wind speed data.If there is low quality data in the real-time monitoring wind speed data,the alarm will be missed or misreported,which seriously affects the operation safety and transportation efficiency of highspeed railway.On the other hand,high winds can cause overturning or derailment of trains in an instant.The safe operation of high-speed railway in a strong wind environment depends on the research on accurate and advanced prediction of instantaneous wind speed at the second level.When there are abnormal data and missing data in historical wind speed data at the second level,it will affect the relevant research of scholars.Therefore,the quality control of wind speed observation data along high-speed railway is very important.Detecting outliers and completing missing values are the key parts of the quality control algorithm.In view of this,this paper counts and analyzes the characteristics of wind speed observation data along the high-speed railway,and constructs a real-time and historical quality control model.The main contents are as follows:(1)In the real-time quality control algorithm of wind speed observation data along the high-speed railway,considering the real-time performance of the algorithm,and in order to avoid false detection and missed detection of a single algorithm,the joint voting mechanism is used to comprehensively consider the detection results of differential threshold method,moving average method and isolated forest method for real-time quality control of data.The experimental results show that the joint voting algorithm has better error detection and accuracy than the three single algorithms,and can effectively reduce the false detection rate and missed detection rate.(2)In the abnormal value detection algorithm of historical wind speed observation data along the high-speed railway,considering the non-stationary and nonlinear characteristics of wind speed observation data,and taking the combined model as the starting point,a quality control model of historical wind speed observation data along the high-speed railway based on LMD-TCN is proposed,so as to provide high-quality data for the research of wind warning along the high-speed railway.The Local Mean Decomposition(LMD)is used to process the original wind speed observation data,and several components are obtained.The Time Convolution Network(TCN)is used to build the model,and the estimated values are finally superposed,and the possible abnormal data are detected by comparing with the original wind speed observation data.The experimental results show that LMD-TCN has high accuracy and error detection.(3)In the missing value completion algorithm of wind speed observation data along highspeed railway,considering that different missing value completion algorithms have different adaptability to missing values of different durations,the missing value compensation is divided into two situations: short-term missing(within 10 s)and long-term missing(outside 10 s)based on the rule of giving an alarm when the wind speed exceeds the alarm threshold for 10 seconds in the overall scheme of Railway Bureau central system of high-speed railway natural disaster and foreign matter intrusion monitoring system(Provisional).Short-time defect completion is fitted by cubic spline interpolation algorithm to obtain the missing value.In the case of longterm missing,the Copula function and Markov chain are used to simulate the actual wind speed,and the final missing completion value is obtained by smoothing the result.The experimental results show that the missing values completed by the algorithm can reflect the fluctuations of actual wind speed observation data.