基于可解释性集成学习的RIOHTrack车辙预测模型及驱动因素研究
doi: 10.21656/1000-0887.450066
edited-by
(我刊编委曹进德来稿)edited-by
详细信息
Research on Driving Factors of the RIOHTrack Rutting Prediction Model Based on Interpretable Ensemble Learning
edited-by
edited-by
(Contributed by CAO Jinde, M.AMM Editorial Board)-
摘要: 交通基础设施是现代社会经济发展的基础，沥青路面作为其中的关键组成部分，扮演着重要角色. 准确预测沥青路面状况对指导路面养护工作具有重要意义. 车辙作为评价沥青路面健康状况的一项重要指标，现有的沥青路面状况预测模型主要基于力学经验模型或机器学习技术. 然而，这些方法缺乏可解释性，无法提供相关信息来说明输入特征对车辙的影响程度. 该研究通过建立可解释性集成学习框架(FI-EL-SHAP)(其中，FI模块通过熵权法和Pareto分析筛选特征，EL模块评估不同的模型性能，并选出最优的模型，SHAP模块对输入特征和模型输出之间的关系进行可视化分析)，揭示了不同特征对模型预测结果的影响. 该研究在保证模型精确度的同时实现了对车辙形成机理的定量解析.Abstract: The transport infrastructure is the foundation of modern social and economic development, where the asphalt pavement plays an important role as a key component. Accurate prediction of asphalt pavement conditions is of great significance to guide pavement maintenance work. Rutting is an important indicator for evaluating the health condition of asphalt pavement. Existing asphalt pavement condition prediction models are mainly based on mechanical experience models or machine learning technologies. However, these methods lack interpretability and cannot provide relevant information on the extent to which the input features affect rutting. Herein, an interpretable integrated learning framework (FI-EL-SHAP) was established, in which the FI module filters features with the entropy weight method and the Pareto analysis, the EL module evaluates the performances of different models and selects the optimal model, and the SHAP module performs visual analysis on the relationship between input features and model outputs to reveal the impacts of different features on model prediction results. This study realizes a quantitative analysis of the rut formation mechanism while ensuring the model accuracy.
-
Key words:
- asphalt pavement /
- rutting /
- ensemble learning /
- interpretability
edited-byedited-by1) (我刊编委曹进德来稿) -
图 3 Pareto分析筛选特征的过程
注 为了解释图中的颜色，读者可以参考本文的电子网页版本，后同.
Figure 3. The process of pareto analysis for feature selection
表 1 STR4数据集
Table 1. The STR4 dataset
axle load average temperature/℃ load level/kN deflection area/mm2 center deflection/(0.01 mm) IRI/(m/km) laser texture depth/mm rut depth/(0.1 mm) 4.45 3.8 56.47 63.75 5.11 1.88 1.01 15.56 5.02 1.2 56.47 64.51 5.34 1.82 1.03 15.85 5.47 0.9 56.47 63.58 5.09 1.87 0.99 14.95 6.92 -0.2 48.09 57.61 4.63 1.93 0.82 35.20 6.96 3.8 48.09 67.59 5.55 1.98 0.89 32.62 ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ 7.74 15.9 48.09 77.68 6.76 1.99 0.90 63.83 
下载: 导出CSV
表 2 不同算法下的STR4的FI
Table 2. The FI of the STR4 under different algorithms
feature RF XGBoost LGBM CatBoost axle load/% 67.84 61.54 49.78 44.55 average temperature/% 4.07 2.12 10.92 10.82 load level/% 0.12 0.00 0.00 4.19 deflection area/% 6.56 2.21 6.55 5.76 center deflection/% 1.95 1.86 3.49 4.92 IRI/% 7.92 4.32 18.78 7.79 laser texture depth/% 11.55 27.96 10.48 21.97
下载: 导出CSV
表 3 熵权法算法权值
Table 3. Entropy weight method algorithm weights
pavement RF XGBoost LGBM CatBoost pavement RF XGBoost LGBM CatBoost STR1 0.40 0.34 0.06 0.20 STR11 0.36 0.39 0.09 0.16 STR2 0.35 0.32 0.12 0.21 STR12 0.31 0.37 0.08 0.24 STR3 0.32 0.38 0.07 0.23 STR13 0.38 0.35 0.07 0.20 STR4 0.28 0.30 0.17 0.25 STR14 0.30 0.41 0.07 0.22 STR5 0.32 0.34 0.12 0.22 STR15 0.37 0.41 0.06 0.16 STR6 0.37 0.28 0.09 0.26 STR16 0.34 0.38 0.08 0.20 STR7 0.34 0.40 0.07 0.19 STR17 0.36 0.37 0.09 0.18 STR8 0.38 0.39 0.07 0.16 STR18 0.37 0.35 0.10 0.18 STR9 0.34 0.35 0.08 0.23 STR19 0.30 0.33 0.10 0.27 STR10 0.37 0.35 0.09 0.19
下载: 导出CSV
表 4 熵权法的FI
Table 4. The FI of the entropy weight method
pavement axle load/% average temperature/% load level/% deflection area/% center deflection/% IRI/% laser texture depth/% STR1 74.64 3.31 4.46 2.50 2.38 3.07 9.63 STR2 56.25 4.39 5.69 1.75 2.33 3.38 26.21 STR3 74.46 3.93 2.04 2.79 2.55 3.17 11.06 STR4 57.04 6.34 1.09 5.05 2.93 8.62 18.93 STR5 72.92 5.38 1.41 3.44 2.39 5.27 9.19 STR6 66.07 5.23 2.51 3.31 3.67 3.56 15.65 STR7 68.51 2.04 5.04 1.64 1.59 3.36 17.82 STR8 80.09 2.94 3.93 2.44 1.37 4.41 4.82 STR9 78.56 4.24 4.25 2.60 2.38 3.88 4.09 STR10 69.67 2.36 3.73 1.70 2.04 2.72 17.78 STR11 81.65 1.79 4.58 2.00 1.07 2.75 6.16 STR12 59.11 2.30 3.18 2.24 2.87 3.88 26.42 STR13 73.02 2.21 3.10 2.12 1.66 4.23 13.66 STR14 30.80 2.43 2.40 1.98 2.04 4.31 56.04 STR15 82.59 1.75 4.82 1.22 1.22 2.36 6.04 STR16 77.58 3.64 3.42 2.15 1.74 2.60 8.87 STR17 79.03 3.44 5.94 2.49 1.56 2.54 5.00 STR18 68.36 2.06 6.97 1.16 1.60 3.46 16.39 STR19 69.38 2.87 1.72 2.98 3.38 5.13 14.54
下载: 导出CSV
表 5 Pareto分析筛选特征
Table 5. Pareto analysis for feature selection
pavement feature STR1, STR2, STR3, STR5, STR6, STR7, STR10, STR12, STR13, STR16, STR18, STR19 axle load, laser texture depth STR4 axle load, laser texture depth, IRI STR8, STR11, STR15 axle load STR9, STR17 axle load, load level STR14 laser texture depth, axle load
下载: 导出CSV
表 6 模型评价指标结果
Table 6. Model evaluation index results
model R2 fRMSE fMAE DF 0.916 4.53 3.42 RF 0.933 4.20 2.96 XGBoost 0.922 4.62 3.33 XGB_RF 0.918 4.83 3.49 CatBoost 0.933 4.22 3.07 LGBM 0.750 8.47 6.31 KNN 0.910 4.80 3.68 SVM 0.491 12.95 9.39 DT 0.894 5.29 3.71 MLP -7.644 51.57 48.32
下载: 导出CSV
-
[1] ZHANG N, ALIPOUR A. A two-level mixed-integer programming model for bridge replacement prioritization[J]. Computer-Aided Civil and Infrastructure Engineering, 2020, 35(2): 116-133. doi: 10.1111/mice.12482 [2] LI Z, ZHANG J, LIU T, et al. Using PSO-SVR algorithm to predict asphalt pavement performance[J]. Journal of Performance of Constructed Facilities, 2021, 35(6): 04021094. doi: 10.1061/(ASCE)CF.1943-5509.0001666 [3] CHOI S, DO M. Development of the road pavement deterioration model based on the deep learning method[J]. Electronics, 2019, 9(1): 3. doi: 10.3390/electronics9010003 [4] DAMIRCHILO F, HOSSEINI A, PARAST M M, et al. Machine learning approach to predict international roughness index using long-term pavement performance data[J]. Journal of Transportation Engineering (Part B): Pavements, 2021, 147(4): 04021058. doi: 10.1061/JPEODX.0000312 [5] HOSSEINI S A, SMADI O. How prediction accuracy can affect the decision-making process in pavement management system[J]. Infrastructures, 2021, 6(2): 28. doi: 10.3390/infrastructures6020028 [6] NASERI H, SHOKOOHI M, JAHANBAKHSH H, et al. Evolutionary and swarm intelligence algorithms on pavement maintenance and rehabilitation planning[J]. International Journal of Pavement Engineering, 2022, 23(13): 4649-4663. doi: 10.1080/10298436.2021.1969019 [7] HOSSAIN E I N, SINGH D, ZAMAN P E M. Dynamic modulus-based field rut prediction model from an instrumented pavement section[J]. Procedia-Social and Behavioral Sciences, 2013, 104: 129-138. doi: 10.1016/j.sbspro.2013.11.105 [8] LI Y, LIU L, XIAO F, et al. Effective temperature for predicting permanent deformation of asphalt pavement[J]. Construction and Building Materials, 2017, 156: 871-879. doi: 10.1016/j.conbuildmat.2017.08.118 [9] 张承烨, 李卓轩, 曹进德. 基于随机k-近邻集成算法的网络流量入侵检测[J]. 南通大学学报(自然科学版), 2023, 22(3): 26-32.ZHANG Chengye, LI Zhuoxuan, CAO Jinde. Network intrusion detection based on random k-nearest neighbor ensemble algorithm[J]. Journal of Nantong University (Natural Science Edition), 2023, 22(3): 26-32. (in Chinese) [10] MIRABDOLAZIMI S M, SHAFABAKHSH G. Rutting depth prediction of hot mix asphalts modified withforta fiber using artificial neural networks and genetic programming technique[J]. Construction and Building Materials, 2017, 148: 666-674. doi: 10.1016/j.conbuildmat.2017.05.088 [11] ZIARI H, AMINI A, GOLI A, et al. Predicting rutting performance of carbon nano tube (CNT) asphalt binders using regression models and neural networks[J]. Construction and Building Materials, 2018, 160: 415-426. doi: 10.1016/j.conbuildmat.2017.11.071 [12] SHAN A, HAFEEZ I, HUSSAN S, et al. Predicting the laboratory rutting response of asphalt mixtures using different neural network algorithms[J]. International Journal of Pavement Engineering, 2022, 23(6): 1948-1956. doi: 10.1080/10298436.2020.1830282 [13] QADIR A, GAZDER U, CHOUDHARY K U N. Artificial neural network models for performance design of asphalt pavements reinforced with geosynthetics[J]. Transportation Research Record: Journal of the Transportation Research Board, 2020, 2674(8): 319-326. doi: 10.1177/0361198120924387 [14] WANG S C. Interdisciplinary Computing in Java Programming[M]. Boston, MA: Springer, 2003. [15] MISHRA M, SRIVASTAVA M. A view of artificial neural network[C]// 2014 International Conference on Advances in Engineering & Technology Research (ICAETR-2014). Unnao, India: IEEE, 2014: 1-3. [16] MAIND S B, WANKAR P. Research paper on basic of artificial neural network[J]. International Journal on Recent and Innovation Trends in Computing and Communication, 2014, 2(1): 96-100. [17] SHANMUGANATHAN S A S. Artificial Neural Network Modelling[M]. Cham: Springer, 2016. [18] SIMPSON A L, DALEIDEN J F, HADLEY W O. Rutting analysis from a different perspective[J]. Transportation Research Record, 1995, 1473: 9-16. [19] SHAFABAKHSH G H, ANI O J, TALEBSAFA M. Artificial neural network modeling (ANN) for predicting rutting performance of nano-modified hot-mix asphalt mixtures containing steel slag aggregates[J]. Construction and Building Materials, 2015, 85: 136-143. doi: 10.1016/j.conbuildmat.2015.03.060 [20] ABDELAZIZ N, ABD EL-HAKIM R T, EL-BADAWY S M, et al. International roughness index prediction model for flexible pavements[J]. International Journal of Pavement Engineering, 2020, 21(1): 88-99. doi: 10.1080/10298436.2018.1441414 [21] BARUA L, ZOU B, NORUZOLIAEE M, et al. A gradient boosting approach to understanding airport runway and taxiway pavement deterioration[J]. International Journal of Pavement Engineering, 2021, 22(13): 1673-1687. doi: 10.1080/10298436.2020.1714616 [22] 王旭东. 从试验环道看长寿命路面的"中国制造"[J]. 中国公路, 2020(14): 30-32. doi: 10.3969/j.issn.1006-3897.2020.14.007WANG Xudong. Viewing the "Made in China" of long life road surface from the experimental ring road[J]. China Highway, 2020(14): 30-32. (in Chinese) doi: 10.3969/j.issn.1006-3897.2020.14.007 [23] 张蕾, 周兴业, 王旭东. 基于RIOHTrack足尺加速加载试验的长寿命沥青路面行为研究进展[J]. 科学通报, 2020, 65(30): 3247-3258.ZHANG Lei, ZHOU Xingye, WANG Xudong. Research progress of long-life asphalt pavement behavior based on the RIOHTrack full-scale accelerated loading test[J]. Chinese Science Bulletin, 2020, 65(30): 3247-3258. (in Chinese) [24] LI Z, SHI X, CAO J, et al. CPSO-XGBoost segmented regression model for asphalt pavement deflection basin area prediction[J]. Science China Technological Sciences, 2022, 65(7): 1470-1481. doi: 10.1007/s11431-021-1972-7 [25] ZEIADA W, HAMAD K, OMAR M, et al. Investigation and modelling of asphalt pavement performance in cold regions[J]. International Journal of Pavement Engineering, 2019, 20(8): 986-997. doi: 10.1080/10298436.2017.1373391 [26] BOMMERT A, SUN X, BISCHL B, et al. Benchmark for filter methods for feature selection in high-dimensional classification data[J]. Computational Statistics & Data Analysis, 2020, 143: 106839. [27] NASERI H, WAYGOOD E O D, WANG B B, et al. How to predict climate change stage of change accurately: proposing a new feature selection technique[C]//Transportation Research Board 101st Annual Meeting. Washington DC, 2022. [28] LIU L, ZHOU J, AN X, et al. Using fuzzy theory and information entropy for water quality assessment in Three Gorges region, China[J]. Expert Systems With Applications, 2010, 37(3): 2517-2521. doi: 10.1016/j.eswa.2009.08.004 [29] ZOU Z H, YUN Y, SUN J N. Entropy method for determination of weight of evaluating indicators in fuzzy synthetic evaluation for water quality assessment[J]. Journal of Environmental Sciences, 2006, 18(5): 1020-1023. doi: 10.1016/S1001-0742(06)60032-6 [30] KIREMIRE A R. The application of the Pareto principle in software engineering[Z/OL]. 2021[2024-11-28]. https://studylib.net/doc/8372157/the-application-of-the-pareto-principle-in-software .[31] REFAEILZADEH P, TANG L, LIU H. Cross-validation[C]//Encyclopedia of Database Systems. Boston, MA: Springer, 2009: 532-538. [32] NGARAMBE J, IRAKOZE A, YUN G Y, et al. Comparative performance of machine learning algorithms in the prediction of indoor daylight illuminances[J]. Sustainability, 2020, 12(11): 4471. doi: 10.3390/su12114471 [33] LUNDBERG S M, LEE S I. A unified approach to interpreting model predictions[C]// 31st Conference on Neural Information Processing Systems. Long Beach, CA, 2017. [34] BREIMAN L. Random forests[J]. Machine Learning, 2001, 45: 5-32. doi: 10.1023/A:1010933404324 [35] CHEN T, GUESTRIN C. XGBoost: a scalable tree boosting system[C]//Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. San Francisco: ACM, 2016: 785-794. [36] PROKHORENKOVA L, GUSEV G, VOROBEV A, et al. CatBoost: unbiased boosting with categorical features[C]//Advances in Neural Information Processing Systems 31. 2018: 6138-6148. [37] KE G L, MENG Q, FINLEY T, et al. LightGBM: a highly efficient gradient boosting decision tree[C]//Advances in Neural Information Processing Systems 30. 2017: 3147-3155. [38] ZHOU Z H, FENG J. Deep forest[J]. National Science Review, 2019, 6(1): 74-86. doi: 10.1093/nsr/nwy108 [39] PETERSON L. K-nearest neighbor[J]. Scholarpedia, 2009, 4(2): 1883. doi: 10.4249/scholarpedia.1883 [40] HEARST M A, DUMAIS S T, OSUNA E, et al. Support vector machines[J]. IEEE Intelligent Systems and Their Applications, 1998, 13(4): 18-28. doi: 10.1109/5254.708428 [41] MYLES A J, FEUDALE R N, LIU Y, et al. An introduction to decision tree modeling[J]. Journal of Chemometrics, 2004, 18(6): 275-285. doi: 10.1002/cem.873 [42] GARDNER M W, DORLING S R. Artificial neural networks (the multilayer perceptron): a review of applications in the atmospheric sciences[J]. Atmospheric Environment, 1998, 32(14/15): 2627-2636. [43] 李卓轩, 林凯迪, 郭建华, 等. 基于车联网数据的运输车辆安全评价模型[J]. 南通大学学报(自然科学版), 2020, 19(1): 26-32.LI Zhuoxuan, LIN Kaidi, GUO Jianhua, et al. Transportation vehicle safety evaluation model based on vehicle network data[J]. Journal of Nantong University (Natural Science Edition), 2020, 19(1): 26-32. (in Chinese) -
图(6) / 表(6)
计量
- 文章访问数: 38
- HTML全文浏览量: 14
- PDF下载量: 3
- 被引次数: 0
渝公网安备50010802005915号