基于Super smoother和3σ原理的列车动态测试趋势性异常数据清洗方法与分析

Method and analysis of train dynamic test trending abnormal data cleaning based on super smoother and 3σ principles

列车动态测试存在数据采集环境干扰大,重复成本高等问题,需要尽可能的从存在异常的数据中保留更多的有效信息。本文针对其中存在的长周期,低频率的趋势性异常数据清洗问题,首先介绍了一种基于Super smoother和3σ原理的数据清洗方法。然后通过与其他常用异常数据清洗方法如神经网络,小波变换等的对比,分别从降噪处理,数据漂移处理,缺失数据补充处理和短暂快速异常波动处理四个方面对方法的数据清洗能力进行了分析和验证,结果表明清洗后数据的Pearson系数由0.785上升到0.923,方法在快速清洗和数据修补方面具有较大优势。最后以某城轨列车制动温升试验数据为例,对实际线路测试数据进行了数据清洗处理,结果表明方法能够较好的解决列车动态测试中存在的趋势性异常数据清洗问题。

Train dynamic testing has the problems such as large data collection environment interference and high repetition cost.It is necessary to retain as much effective information as possible from abnormal data.Aiming at the long-period,low-frequency trending abnormal data cleaning problem,this article first introduces a data cleaning method based on Super smoother and 3σprinciples.Then,through comparing with other commonly used abnormal data cleaning methods such as neural network,wavelet transform,etc.,the data cleaning ability of the method is analyzed and verified from four aspects:noise reduction processing,data drift processing,missing data supplementing processing and transient and rapid abnormal fluctuation processing.Analysis and verification results show that the Pearson coefficient of the data after cleaning is risen from 0.785 to 0.923.The method has great advantages in fast cleaning and data repair.Finally,the brake temperature rising test data of a city rail train was taken as an example,the actual line test data were processed for data cleaning.The results show that the method can better solve the trending abnormal data cleaning problem existing in the train dynamic test.