Research on Temperature Model of Steel Box Girder of High-Speed Railway Cable-Stayed Bridge Based on Machine Learning

doi:10.12141/j.issn.1000-565X.240200

Abstract

Abstract:

To investigate the temperature patterns of steel box girders in long-span cable-stayed bridges on high-speed railways, this study utilized measured temperature data from the Yuxi River Bridge on the Shangqiu-Hefei-Hangzhou High-Speed Railway, along with database resources. By employing machine learning techniques, the research explored the influence of various meteorological factors on the temperature behavior of steel box girders, as well as the temporal and spatial distribution characteristics of the temperature field. By establishing machine learning mo-dels that map various meteorological factors to the uniform temperature of the steel box girder, the superiority, inferio-rity, and applicability of each model were analyzed, and the importance ranking of meteorological factors affecting the uniform temperature of the steel box girder was obtained. A comprehensive study on the vertical distribution pattern of the temperature of the steel box girder was conducted using machine learning methods and exponential fitting. The results show that the importance ranking of meteorological factors affecting the uniform temperature of the steel box girder from high to low is: air temperature, cumulative radiation, air pressure, humidity, radiation intensity, wind direction, horizontal visibility, wind speed, and precipitation, with the temperature importance far exceeding other meteorological factors. Among them, the atmospheric temperature 2 to 3 hours ago has the greatest impact on the uniform temperature of the steel box girder, reflecting a lag of 2 to 3 hours in the impact of atmospheric temperature changes on the uniform temperature of the steel box girder. Neural networks, random forests, and XGBoost models can all accurately predict the uniform temperature of the steel box girder, with the neural network model performing better overall. The negative temperature gradient in the steel box girder exhibits lower sensitivity to meteorological factors and is more strongly correlated with the internal heat transfer characteristics of the structure itself. The exponential function can accurately fit the vertical distribution of the maximum positive temperature gradient in steel box girders, with its parameters determinable through machine learning methods. Each parameter holds distinct physical significance. The research findings provide valuable reference for predicting temperature fields and understanding distribution patterns in the steel box girders of long-span cable-stayed bridges on high-speed railways.

Key words: high-speed railway, steel box girder, temperature mode, machine learning, meteorological factor

CLC Number:

U238

LIU Wenshuo, ZHONG Mingfeng, ZHOU Bo, LÜ Fangzhou. Research on Temperature Model of Steel Box Girder of High-Speed Railway Cable-Stayed Bridge Based on Machine Learning[J]. Journal of South China University of Technology(Natural Science Edition), 2025, 53(6): 25-33.

Figures/Tables 16

Fig.1

Fig.2

Fig.3

Table 1

Table 2

Fig.4

Table 3

Table 4

Table 5

Fig.5

Fig.6

Table 6

Fig.7

Fig.8

Fig.9

Fig.10

References 20

1	缪长青，史长华．大跨悬索桥扁平钢箱梁温度梯度与温度影响研究［J］．中国科学：技术科学，2013，43（10）：1155-1164.
	MIAO Changqing， SHI Changhua ．Temperature gradient and its effect on flat steel box girder of long-span suspension bridge［J］．Science China：Technological Science，2013，43（10）：1155-1164.
2	王高新，丁幼亮，王晓晶，等．苏通大桥扁平钢箱梁温度场长期监测与统计分析［J］．公路交通科技，2014，31（2）：69-73，120.
	WANG Gao-xin， DING You-liang， WANG Xiao-jing，et al ．Long-term monitoring and statistical analysis of temperature field of flat steel-box girder of Sutong Bridge［J］．Journal of Highway Transportation Research and Development，2014，31（2）：69-73，120.
3	许翔，黄侨，任远，等．大跨钢斜拉桥实测结构温度场分析［J］．哈尔滨工业大学学报，2019，51（9）：14-21.
	XU Xiang， HUANG Qiao， REN Yuan，et al ．Thermal field analysis for large span steel cable-stayed bridges using in-situ measurements［J］．Journal of Harbin Institute of Technology，2019，51（9）：14-21.
4	HUANG X， ZHU J， LI Y L ．Temperature analysis of steel box girder considering actual wind field［J］．Engineering Structures， 2021， 246： 113020/1-17.
5	滕华俊，祝志文，李健朋．基于现场实测的钢桥面钢箱梁竖向温度梯度研究［J］．铁道科学与工程学报，2021，18（12）：3267-3277.
	TENG Huajun， WU Zhiwen， LI Jianpeng ．Investigation on vertical temperature gradient of steel box girder with steel deck under solar radiation［J］．Journal of Railway Science and Engineering，2021，18（12）：3267-3277.
6	LU H L， HAO J， ZHONG J W，et al ．Analysis of sunshine temperature field of steel box girder based on monitoring data［J］．Advances in Civil Engineering，2020，2020（1）： 9572602/1-10.
7	黎泰良．基于ANSYS的无铺装轨道钢箱梁日照温度场敏感性分析［J］．工程建设与设计，2022，471（1）：56-60.
	LI Tai-liang ．Sensitivity analysis of temperature field of unpaved track steel box girder based on ANSYS［J］．Construction & Design for Engineering，2022，471（1）：56-60.
8	章李刚，陈仲江，吴博，等．跨海大桥钢箱梁温度场特性分析及降温措施研究［J］．重庆交通大学学报（自然科学版），2023，42（1）：9-14.
	ZHANG Ligang， CHEN Zhongjiang， WU Bo，et al ．Temperature field characteristics and cooling measures of steel box girder of cross-sea bridge［J］．Journal of Chongqing Jiaotong University （Natural Science），2023，42（1）： 9-14.
9	张钾奇．太阳辐射下钢箱梁温度效应的多尺度有限元模型研究［D］．武汉：武汉工程大学，2023.
10	张洁．基于气象数据驱动和深度学习的桥梁温度和应变预测［Ｄ］．广州：广州大学，2023.
11	PEARSON E S， SNOW B A S ．Tests for rank correlation coefficients Ⅲ．Distribution of the transformed kendall coefficient［J］．Biometrika，1962，49（1）：185-191.
12	周志华．机器学习［M］．北京：清华大学出版社，2016.
13	王德明，王莉，张广明．基于遗传BP神经网络的短期风速预测模型［J］．浙江大学学报（工学版），2012，46（5）：837-841，904.
	WANG De-ming， WANG Li， ZHANG Guang-ming ．Short-term wind speed forecast model for wind farms based on genetic BP neural network［J］．Journal of Zhejiang University （Engineering Edition），2012，46（5）：837-841，904.
14	王凯，张勇，刘建磊，等．基于BP神经网络的混凝土箱梁最大温度梯度预测［J］．铁道科学与工程学报，2024，21（2）：837-850.
	WANG Kai， ZHANG Yong， LIU Jianlei，et al ．Prediction of concrete box-girder maximum temperature gradient based on BP neural network［J］．Journal of Railway Science and Engineering，2024，21（2）：837-850.
15	WANG K W， REN J， YAN J W，et al ．Research on a concrete compressive strength prediction method based on the random forest and LCSSA-improved BP neural network［J］．Journal of Building Engineering，2023，76：107150/1-15.
16	曹正凤．随机森林算法优化研究［D］．北京：首都经济贸易大学，2014.
17	陶晔，杜景林．基于随机森林的长短期记忆网络气温预测［J］．计算机工程与设计，2019，40（3）：737-743.
	TAO Ye， DU Jing-lin ．Temperature prediction using long short term memory network based on random forest［J］．Computer Engineering and Design，2019，40（3）：737-743.
18	SUN Z G， WANG G T， LI P F，et al ．An improved random forest based on the classification accuracy and correlation measurement of decision trees［J］．Expert Systems with Applications，2024，237：121549/1-19.
19	CHEN T Q， GUESTRIN C ．XGBoost：a scalable tree boosting system［C］∥Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Disco-very and Data Mining．San Francisco，CA：［s.n.］，2016： 785-794.
20	钱宇，何益丰，谢斌鑫，等．基于特征工程的XGboost风速短期预测［J］．武汉大学学报（工学版），2022，55（10）：1057-1064.
	QIAN Yu， HE Yifeng， XIE Binxin，et al ．Short-term XGboost wind speed prediction based on feature enginee-ring［J］．Engineering Journal of Wuhan University，2022，55（10）：1057-1064.

气象特征	气压	最高气压	最低气压	海平面气压
气压	1.000	0.999	0.999	0.999
最高气压	0.999	1.000	0.999	0.999
最低气压	0.999	0.999	1.000	0.999
海平面气压	0.999	0.999	0.999	1.000

气象特征	相关系数		相关系数
气压	-0.898	辐射强度I₅	0.429
海平面气压	-0.898	辐射强度I₆	0.408
最高气压	-0.898	辐射强度I₇	0.375
最低气压	-0.898	辐射强度I₈	0.336
最大风速	0.187	辐射强度I₉	0.294
实时风速	0.143	辐射量E₃	0.366
最大风速风向	-0.181	辐射量E₄	0.396
实时风向	-0.170	辐射量E₆	0.452
最高气温	0.976	辐射量E₇	0.476
最低气温	0.975	辐射量E₈	0.497
相对湿度	-0.248	辐射量E₉	0.515
水汽压	0.893	辐射量E₁₀	0.531
降水量	0.022	辐射量E₁₂	0.559
水平能见度	0.562	辐射量E₁₃	0.570
气温θ₀	0.972	辐射量E₁₄	0.581
气温θ₁	0.979	辐射量E₁₅	0.591
气温θ₂	0.982	辐射量E₁₆	0.602
气温θ₃	0.981	辐射量E₁₇	0.614
辐射强度I₀	0.301	辐射量E₂₁	0.667
辐射强度I₁	0.356	辐射量E₂₂	0.680
辐射强度I₂	0.400	辐射量E₂₃	0.690
辐射强度I₃	0.429	辐射量E₂₄	0.699

超参数	取值
激活函数	Tanh
优化器	L-BFGS
学习率	0.1
L2正则项	1.0
丢弃率	0
训练步数	10 000

超参数	取值
节点分裂评价准则	MSE（均方误差）
划分时考虑的最大特征比例	无
内部节点分裂的最小样本数	2
叶子节点的最小样本数	1
叶子节点中样本的最小权重	0
树的最大深度	50
叶子节点的最大数量	500
决策树数量	500
是否有放回采样	是

超参数	取值
基学习器	gbtree（梯度提升树）
基学习器数量	1 000
学习率	0.1
L1正则项	0
L2正则项	1
样本特征采样率	1
树特征采样率	1
节点特征采样率	1
叶子节点中样本的最小权重	0
树的最大深度	1 000