基于BP神经网络的铁路轨道几何不平顺预测方法

Prediction Method of Railway Track Geometric Irregularity Based on BP Neural Network

设备密集型重载铁路对轨道平顺状态预测技术十分重视,受限于数据挖掘分析技术,轨道检测车在养护维修计划决策支持方面还未完全发挥应有的作用。本文根据轨道不平顺的变化特点,采用神经网络方法对重载铁路轨道不平顺7项参数进行预测,从而为养护维修策略的决策提供支持。将某重载铁路K420+000~K426+000区段长达18个月的轨道不平顺检测数据用于模型训练并进行预测分析,结果显示:双隐层、单隐层BP网络模型和多元回归分析模型的均方预测误差平均值分别为0.065 2、0.068 9、0.105 1,平均相对误差分别为8.03%、8.65%、11.57%。双隐层BP网络模型模型精度更高,该模型为重载铁路轨道不平顺发展预测的研究提供了新的思路。

Prediction technique of track irregularity is crucial to equipment-intensive heavy haul railway.Limited to the data mining analysis technology,track inspection car still could not fully play its role in decision support for the maintenance plan.Based on the variation characteristics of track irregularity,this paper used the neural network method to predict 7 track irregularity parameters of a heavy haul railway,so as to provide support for the policy-making of maintenance strategy.The track irregularity data of a heavy load railway at K420+000~K426+000 section for 18 months were used for model training and prediction analysis.The results show that the mean square prediction error of the double hidden layer,the single hidden layer BP network model and the multiple regression analysis model is 0.065 2,0.068 9,0.105 1 respectively,and the average relative error is 8.03%,8.65%,11.57%respectively.The double hidden layer BP network model,with higher accuracy,provides a new way of thinking for the development prediction of heavy haul railway track irregularities.