随机森林算法在钢轨闪光焊接工艺质量预测模型数据预分类的应用  

Application of Pre-classification of Data in Quality Prediction Model for Rail Flash Butt Welding Process Using Random Forest Algorithm

作　　者：周烨 ZHOU Ye(Metals and Chemistry Research Institute,China Academy of Railway Sciences Corporation Limited,Beijing 100081,China)

机构地区：[1]中国铁道科学研究院集团有限公司金属及化学研究所,北京100081

出　　处：《高速铁路新材料》2024年第3期63-67,共5页Advanced Materials of High Speed Railway

基　　金：中国铁道科学研究院集团有限公司科研项目(2023YJ045)。

摘　　要：针对训练钢轨闪光焊接头质量预测深度学习模型需要大量数据集且采集困难的问题,采用预分类的方法,只选取相同需求顶锻量范围内的数据进行训练,以降低训练集数量并提高数据质量。以钢轨闪光焊接工艺的预测顶锻量为研究对象,随机森林回归算法进行建模,采用13个关键工艺参数预测实际顶锻量。研究结果表明:预测模型精度明显优于传统的理论建模及回归分析。在比较了多种常规算法的预测模型后发现,在训练样本较少的情况下,随机森林回归算法模型表现最佳。进一步数据分析得出,焊接工艺参数中的中期脉动阶段时间值、闪光末期钢轨消耗量和中期脉动钢轨消耗量对实际顶锻量影响权重相对较大。基于随机森林回归算法的预测模型为焊接实际需求的顶锻量预测及控制提供了一种有效的方法。In addressing the challenge of acquiring a substantial dataset for predicting the quality of flash butt welding joints in rail training,a pre-classification approach is adopted.This method selectively trains on data within the same required forging quantity range to diminish the demand for training set size and enhance data quality.Our focus lies in predicting the forging quantity in flash butt welding processes of rails,for which we employ a random forest regression algorithm for modeling.Our objective is to forecast the actual forging quantity while considering 13 key acquisition parameters in the process.Our research findings demonstrate that the predictive model´s accuracy significantly surpasses that of traditional theoretical modeling and regression analysis.Comparative analysis among various conventional algorithmic predictive models reveals the superior performance of the random forest regression model,particularly when training samples are limited.Further data analysis indicates that parameters of the welding process,such as the time value of the mid-welding phase,flash end consumption,and midwelding consumption,exert relatively significant influences on the actual forging quantity.The predictive model based on the random forest regression algorithm furnishes an effective means for forecasting and regulating the forging quantity essential for welding applications.

关 键 词：钢轨闪光焊接 工艺 顶锻量预测 随机森林算法 

分 类 号：U213.92[交通运输工程—道路与铁道工程]