| 66 | 0 | 48 |
| 下载次数 | 被引频次 | 阅读次数 |
现有铁路轨道几何预测方法无法完全考虑数据中的趋势性、季节性和随机性，限制了长期预测的精度。本文提出了一种基于STL分解、融合ARIMA与SARIMA的组合模型，用于轨道几何状态的长期预测。首先，利用五年的轨道几何状态检测数据以及同期气候资料，分析轨道几何的退化规律和季节波动特性，并通过傅里叶变换确定了其变化周期。而后，使用STL算法将原始序列分解为趋势、季节和残差子序列，对每个子序列分别使用ARIMA和SARIMA模型进行建模，并选择最优模型。最后，通过STL逆分解重新组合最优模型对每个子序列的预测结果，得到轨道几何数据的整体预测值。以某重载铁路实测数据进行验证，结果表明：组合模型在直线区段的预测均方误差(MSE)为0.0233，平均绝对误差(MAE)为0.1205，决定系数(R2)为0.90；在曲线区段的MSE为0.0029，MAE为0.0471，R2为0.91，均明显优于单一ARIMA和SARIMA模型。该方法能有效适应轨道几何数据的复杂波动，具有良好的长期预测性能和工程应用价值。
Abstract：Existing railway track geometry prediction methods do not fully account for trends， seasonality， and randomness in the data， limiting the accuracy of long-term predictions. This study proposes a combined model based on STL decomposition that integrates ARIMA and SARIMA for long-term prediction of track geometry conditions. First， five years of track geometry inspection data and corresponding climate records are analyzed to characterize track geometry degradation and seasonal fluctuations， and Fourier transform is used to identify the dominant period. The STL algorithm is then applied to decompose the original series into trend， seasonal， and residual components. ARIMA and SARIMA models are fitted to each component， and the better-performing model is selected for prediction. Finally， the component forecasts are recombined through additive reconstruction to obtain the overall track geometry predictions. Validation using field measurements from a heavy-haul railway shows that the combined model achieves a mean squared error (MSE) of 0.0233， a mean absolute error (MAE) of 0.1205， and a coefficient of determination (R2) of 0.90 for the straight track section. For the curved track section， the corresponding values are 0.0029， 0.0471， and 0.91， respectively. The combined model consistently outperforms the individual ARIMA and SARIMA models across these metrics. The proposed method effectively accommodates complex fluctuations in track geometry data and demonstrates strong long-term prediction performance and potential for engineering applications.
[1] Jovanović S， Guler H， Čoko B. Track degradation analysis in the scope of railway infrastructure maintenance management systems[J]. Gradjevinar， 2015， 67(3): 247-258.
[2] GUO Ran， HAN Baoming. Multi-stage linear prediction model for railway track irregularity[J]. Applied mechanics and materials， 2013， 361-363: 1811-1816.
[3] JIA Chaolong， WEI Lili， WANG Hanning. et al. A hybrid model based on wavelet decomposition-reconstruction in track irregularity state forecasting[J]. Mathematical problems in engineering， 2015， 2015: 548720.
[4] ZHOU Ge. Railway track irregularity data mining and time series trend forecasting research[J]. Applied mechanics and materials， 2014， 666: 272-275.
[5] 林怀青，柴晓冬，苏圣超，等.基于PCA和SVM的轨道不平顺状态识别[J].测控技术，2016，35(5):25-28.
[6] 许玉德，刘一鸣，沈坚锋.轨道不平顺预测随机模型的SVM-MC求解方法[J].华东交通大学学报，2018，35(3):1-7.
[7] 彭丽宇，张进川，苟娟琼，等.基于BP神经网络的铁路轨道几何不平顺预测方法[J].铁道学报，2018，40(9):154-158.
[8] 韩磊.轨道几何不平顺预测中机器学习的应用研究[D].石家庄:石家庄铁道大学，2023:34-50.
[9] HAN Lei， LIAO Yingying， WANG Haoyu， et al. Analysis and prediction of railway track longitudinal level using multiple machine learning methods[J]. Measurement science and technology， 2024， 35(2): 24001.
[10] WANG Xin， BAI Yun， LIU Xiang. Prediction of railroad track geometry change using a hybrid CNN-LSTM spatial-temporal model[J]. Advanced Engineering Informatics， 2023， 58: 102235.
[11] Theodosiou M. Forecasting monthly and quarterly time series using STL decomposition[J]. International Journal of Forecasting， 2011， 27(4): 1178-1195.
[12] ZHAO Jiandong， YU Zhixin， YANG Xin， et al. Short term traffic flow prediction of expressway service area based on STL-OMS[J]. Physica a: statistical mechanics and its applications， 2022， 595: 126937.
基本信息:
中图分类号:U216.3
引用信息:
[1]韩磊,张喆,段姝琪,等.考虑气候季节性因素的轨道几何长期预测组合模型[J].铁道建筑().
基金信息:
内蒙古自治区自然科学基金（2022QN05033）
2026-09-11
2026-09-11
2026-09-11