一种基于GBRT算法的CA砂浆脱空检测方法-铁道科学与工程学报.PDFVIP

第 15 卷 第 2 期 铁道科学与工程学报 Volume 15 Number 2 2018 年 2 月 Journal of Railway Science and Engineering February 2018 一种基于 GBRT 算法的 CA 砂浆 脱空检测方法 李自法，谢维波，刘涛 (华侨大学 计算机科学与技术学院，福建 厦门 361021) 摘 要：不同于现有检测方法，利用 CA 砂浆脱空和非脱空情况下采集到的声音信号不同，提出把机器学习领域中的 GBRT 算法应用到 CA 砂浆脱空检测领域。利用脱空和非脱空情况下采集到的数据，使用 GBRT 算法训练出一个二分类模型；通过 对输入数据做类别决策，判断对应的采集位置是否脱空。介绍算法原理，并结合 CA 砂浆脱空问题，分析模型的使用技巧； 引入其他主流机器学习分类算法进行比较分析。实验结果证明，GBRT 算法构造的模型性能较好，在 CA 砂浆脱空检测领域 具有应用前景。 关键词：无砟轨道；CA 砂浆；脱空；GBRT 算法；机器学习分类算法 中图分类号：U213.2 文献标志码：A 文章编号：1672−7029(2018)02 −0292 −10 A detection method of CA mortar disengaging based on GBRT algorithm LI Zifa, XIE Weibo, LIU Tao (School of Computer Science and Technology, Huaqiao University, Xiamen 361021, China) Abstract: Different from the existing detection methods, the GBRT algorithm of machine learning was applied in the CA mortar disengaging detection field using the difference between the sound signal of track slab collected at void and non-void situation for the first time. Using the data collected at the void and non-void situation, a binary classification model was trained with the GBRT algorithm. Category-decision was made for the input data and whether corresponding gathering place is void was judged. The algorithm principle was explained in detail, and the use skills of the model were analyzed combined with the CA mortar disengaging detection. Other mainstream machine learning classification algorithms were introduced for comparative analysis. And the result shows that the GBRT algorithm is much better and has a great application prospect in this field. Key words: ballastless track; CA mortar; disengaging; GBRT algorithm; machine learnin