摘要
车轮踏面磨耗量是评价机车运行安全的重要参数,但多数车轮运维现场的条件尚不能对踏面磨耗量及时、准确地进行监测。为了解决上述问题,文章提出了基于“GA-岭回归”分析的机车车轮踏面磨耗量预测算法(简称“GA-岭回归”预测算法)。“GA-岭回归”预测算法分为数据前处理和数据预测分析2部分。对于数据前处理,首先依据不同的测量方式对采集到的踏面磨耗量数据进行分类,基于车轮的实际运维情况,分析不同类型数据的特点;随后以镟修周期作为数据划分标准,对分类后的数据进行切片处理;最后采用相关标准和主成分分析法对相应的动态测量数据进行清洗、降噪。对于数据预测分析,首先划分数据集,进行数据整合,创建训练集数据的时间滑动窗口;随后采用岭回归算法对训练集数据进行回归分析训练,并结合遗传算法和验证集数据进行模型参数的调优,以提高预测准确性。基于测试集数据,分别采用传统预测算法、岭回归线性预测算法和“GA-岭回归”预测算法,对3类预测算法的预测效果进行对比分析;随后采用相同的测试方法,扩大车轮样本的数量,继续进行预测效果的对比分析。测试结果表明,采用“GA-岭回归”预测算法的预测误差和误差标准差均相对较低。经分析可知,“GA-岭回归”预测算法具有较高的预测精度,同时可保证更好的预测稳定性。
Wheel tread wear is an important parameter to evaluate the operational safety of locomotives,yet timely and accurate monitoring is often lacking at wheel operation and maintenance sites.To this end,this paper proposed a prediction algorithm of tread wear for locomotive wheels based on GA-ridge regression analysis(hereinafter referred to as the"GA-ridge regression"prediction algorithm).This algorithm consisted of two steps:data pre-processing and data-based prediction analysis.In the first step,collected tread wear data was classified according to different measurement methods,and characteristics of different data types were analyzed considering the actual operation and maintenance of wheels.The classified data was then sliced using the profiling cycle as the data partitioning criterion,followed by cleaning and noise reduction of the corresponding dynamic measurement data by relevant criteria and principal component analysis.In the second step,data was integrated into datasets,and a time-sliding window was created for the training set data.The ridge regression algorithm was used to train the training set data for regression analysis,and the model parameters were tuned using a combination of the genetic algorithm and the validation set data to improve the prediction accuracy.The test set data was used for prediction by the traditional prediction algorithm,ridge regression linear prediction algorithm,and GA-ridge regression prediction algorithm respectively to compare and analyze their prediction effects.Additionally,a comparative analysis was conducted using the same evaluation method and sample wheels in an expanded size to further assess the prediction effects.The results indicate relatively lower prediction errors and standard deviations of errors when using the GA-ridge regression prediction algorithm.This research concludes that the GA-ridge regression prediction algorithm provides higher prediction accuracy and better prediction stability.
作者
方鑫
刘通
程亚萍
孙宇铎
王菲儿
FANG Xin;LIU Tong;CHENG Yaping;SUN Yuduo;WANG Feier(Shanghai Depot,China Railway Shanghai Group Co.,Ltd.,Shanghai 200070,China;Metals and Chemistry Research Institute,China Academy of Railway Sciences Corporation Limited,Beijing 100081,China)
出处
《机车电传动》
北大核心
2023年第6期71-78,共8页
Electric Drive for Locomotives
基金
中国国家铁路集团有限公司科技研究开发计划项目(P2023J001)
中国铁道科学研究院集团有限公司科研项目(2022YJ245)。
关键词
机车车轮
车轮踏面磨耗量预测
主成分分析
岭回归
遗传算法
locomotive wheels
wheel tread wear prediction
principal component analysis
ridge regression
genetic algorithm
