Prediction model for displacement data of railway infrastructure
-
摘要: 线路安全是铁路运营的重要前提,我国铁路跨度广、行车环境复杂,当铁路基础设施稳定性产生改变时往往会严重影响行车安全。文章采用长短期记忆(LSTM ,Long Short-Term Memory)模型对基于全球导航卫星系统(GNSS,Global Navigation Satellite System)的铁路基础设施监测系统的形变监测数据进行建模预测,实现对铁路基础设施灾害的早期预警,并与多种传统时间序列预测模型进行对比,结果表明,LSTM模型具有更好的性能。Abstract: Railway safety is an important prerequisite for operation. Due to the wide span and complex driving environment of China's railway, when the stability of railway infrastructure changes, it will seriously affect the driving safety. This paper used Long Short-Term Memory(LSTM)model to model and predict the deformation monitoring data of railway infrastructure monitoring system based on Global Navigation Satellite System(GNSS), Achieved early warning of railway infrastructure disasters, and compared it with various traditional time series prediction models. The experimental results show that the LSTM model has better performance.
-
-
表 1 发育型监测数据预测精度指标
东西 南北 垂直 Time RMSE MAE Time RMSE MAE Time RMSE MAE DT 0.04 0.101 0.083 0.05 0.045 0.033 0.08 0.044 0.034 Linear 0.06 0.011 0.008 0.06 0.026 0.019 0.08 0.017 0.013 SVM 0.08 0.240 0.212 0.08 0.073 0.042 0.017 0.099 0.067 RF 0.11 0.099 0.079 0.12 0.037 0.026 0.10 0.032 0.024 GBRT 0.16 0.103 0.083 0.16 0.038 0.025 0.08 0.033 0.026 ET 0.19 0.102 0.084 0.21 0.045 0.033 0.10 0.051 0.042 ARD 0.21 0.011 0.008 0.23 0.026 0.019 0.44 0.017 0.013 Byesian 0.23 0.011 0.008 0.25 0.026 0.019 0.08 0.017 0.013 TheilSen 0.25 0.011 0.008 0.27 0.026 0.019 0.12 0.016 0.012 RANSAC 0.28 0.011 0.008 0.29 0.026 0.019 0.28 0.017 0.012 LSTM8 2.55 0.030 0.023 2.55 0.025 0.017 2.55 0.018 0.013 LSTM12 2.60 0.035 0.026 2.55 0.027 0.019 2.60 0.021 0.015 表 2 稳定型监测数据预测精度指标
东西 南北 垂直 Time RMSE MAE Time RMSE MAE Time RMSE MAE DT 0.04 0.060 0.048 0.05 0.048 0.035 0.04 0.052 0.042 Linear 0.06 0.041 0.033 0.06 0.030 0.023 0.06 0.031 0.024 SVM 0.07 0.068 0.053 0.08 0.053 0.040 0.07 0.091 0.066 RF 0.11 0.049 0.039 0.12 0.034 0.028 0.10 0.045 0.034 GBRT 0.14 0.048 0.038 0.17 0.035 0.028 0.14 0.046 0.035 ET 0.17 0.066 0.052 0.21 0.055 0.041 0.17 0.056 0.045 ARD 0.19 0.041 0.033 0.23 0.030 0.023 0.19 0.031 0.025 Byesian 0.20 0.041 0.033 0.25 0.030 0.023 0.21 0.031 0.024 TheilSen 0.22 0.047 0.039 0.27 0.031 0.024 0.22 0.031 0.025 RANSAC 0.23 0.046 0.038 0.29 0.031 0.024 0.24 0.031 0.024 LSTM8 2.55 0.038 0.031 2.58 0.034 0.027 2.59 0.031 0.024 LSTM12 2.53 0.040 0.032 2.60 0.027 0.021 2.60 0.030 0.023 表 3 波动稳定型监测数据预测精度指标
东西 南北 垂直 Time RMSE MAE Time RMSE MAE Time RMSE MAE DT 0.04 0.084 0.067 0.05 0.067 0.053 0.04 0.067 0.056 Linear 0.06 0.049 0.038 0.08 0.040 0.031 0.06 0.048 0.039 SVM 0.08 0.064 0.047 0.10 0.080 0.051 0.08 0.066 0.053 RF 0.11 0.060 0.047 0.13 0.050 0.039 0.11 0.054 0.044 GBRT 0.15 0.054 0.043 0.17 0.052 0.040 0.16 0.055 0.043 ET 0.18 0.080 0.064 0.21 0.073 0.056 0.19 0.078 0.064 ARD 0.19 0.049 0.038 0.22 0.040 0.031 0.20 0.047 0.039 Byesian 0.21 0.049 0.037 0.24 0.040 0.031 0.22 0.048 0.039 TheilSen 0.22 0.052 0.040 0.26 0.042 0.033 0.24 0.049 0.040 RANSAC 0.24 0.052 0.040 0.27 0.041 0.031 0.25 0.051 0.041 LSTM8 2.68 0.041 0.032 2.46 0.036 0.028 2.76 0.046 0.037 LSTM12 2.60 0.043 0.034 2.62 0.035 0.026 2.65 0.048 0.038 -
[1] 雷绍兰,孙才新,周 湶. 电力短期负荷的多变量时间序列线性回归预测方法研究 [J]. 中国电机工程学报,2006(2):25-29. DOI: 10.3321/j.issn:0258-8013.2006.02.005 [2] 刘 静,关 伟. 交通流预测方法综述 [J]. 公路交通科技,2004(3):82-85. DOI: 10.3969/j.issn.1002-0268.2004.03.022 [3] 杨秀媛,肖 洋,陈树勇. 风电场风速和发电功率预测研究 [J]. 中国电机工程学报,2005(11):1-5. DOI: 10.3321/j.issn:0258-8013.2005.11.001 [4] CHO K, VAN MERRIëNBOER B, GULCEHRE C, et al. Learning phrase representations using RNN encoder-decoder for statistical machine translation [J]. arXiv preprint arXiv:, 1406, 1078: 2014.
[5] WERBOS P J. Generalization of backpropagation with application to a recurrent gas market model [J]. Neural networks, 1988, 1(4): 339-356. DOI: 10.1016/0893-6080(88)90007-X
[6] HOCHREITER S. Untersuchungen zu dynamischen neuronalen Netzen [J]. Diploma, Technische Universität München, 1991, 91(1).
[7] HOCHREITER S, SCHMIDHUBER J. Long short-term memory [J]. Neural computation, 1997, 9(8): 1735-1780. DOI: 10.1162/neco.1997.9.8.1735
[8] GERS F A, SCHMIDHUBER J. LSTM recurrent networks learn simple context-free and context-sensitive languages [J]. IEEE transactions on neural networks, 2001, 12(6): 1333-1340. DOI: 10.1109/72.963769
[9] Srivastava N , Mansimov E , Salakhutdinov R . Unsupervised Learning of Video Representations using LSTMs[J]. JMLR. org, 2015.
[10] HANSON J, YANG Y, PALIWAL K, et al. Improving protein disorder prediction by deep bidirectional long short-term memory recurrent neural networks [J]. Bioinformatics, 2017, 33(5): 685-692.
[11] ECK D, LAPALME J. Learning musical structure directly from sequences of music [J]. University of Montreal, Department of Computer Science, 2008.
[12] GREFF K, SRIVASTAVA R K, KOUTNíK J, et al. LSTM: A search space odyssey [J]. IEEE transactions on neural networks and learning systems, 2016, 28(10): 2222-2232.
-
期刊类型引用(3)
1. 田朋溢,石雯,陈彪,高攀,蒋成成,刘鹏. 无线分布式轨道车辆平稳性测试技术研究. 铁道机车车辆. 2024(06): 110-116 . 百度学术
2. 茅迿,高琦,梁师嵩. 城轨列车智能综合检测系统的研究与应用. 机车车辆工艺. 2023(01): 12-16 . 百度学术
3. 刘国联,陈根,张敏海. 基于CAN总线的铁路道岔监测系统设计. 南方农机. 2020(09): 217-219 . 百度学术
其他类型引用(1)