Prediction model for performance decay of airport runways based on CNN-xLSTM-Attention
Article Text (Baidu Translation)
-
摘要: 为精准预估复杂工况下机场跑道的性能衰变趋势，基于改进深度学习架构建立了CNN-xLSTM-Attention组合预测模型。利用卷积神经网络提取环境、荷载及结构等多源输入数据的局部特征，通过引入包含sLSTM与mLSTM双分支的xLSTM模型以增强对时序依赖的捕捉能力；结合Attention机制动态聚焦关键时间步与特征变量，并采用贝叶斯优化算法自动搜索关键超参数，构建了高精度的PCI衰变预测框架；采用中国两跑道实测时序数据，涵盖通航时间、降雨量、交通荷载等6类关键指标，计算分析了不同模型的预测精度与泛化能力，并引入SHAP与广义加性模型解析了关键特征的非线性影响机制。研究结果表明：相较于传统LSTM及单一改进模型，CNN-xLSTM-Attention模型具有最优的综合性能，测试集决定系数达到0.922 5，平均绝对百分比误差仅为1.777 8%，有效解决了长序列梯度问题与特征耦合难题；解释性分析发现，通航时间（贡献率为44.8%）与年平均降雨量（贡献率为33.6%）是主导道面状况指数（PCI）衰减的首要因素，交通荷载与温度指标存在显著的阈值效应，而增加面层厚度对延缓衰变具有显著抑制作用。由此可见，建立的道面性能预测模型及其解释性分析方法能够准确捕捉PCI的时序演化规律与关键影响机制，可为机场跑道的分阶段科学养护与运维决策提供可靠的数据支撑。Abstract: To accurately predict the performance decay trend of airport runways under complex working conditions, a combined CNN-xLSTM-Attention prediction model was established based on an improved deep learning architecture. A convolutional neural network was utilized to extract local features of multi-source input data such as environment, load, and structure, and an xLSTM model containing dual branches of sLSTM and mLSTM was introduced to enhance the capturing ability for temporal dependencies. On this basis, an Attention mechanism was combined to dynamically focus on key time steps and feature variables, and a Bayesian optimization algorithm was adopted to automatically search for key hyperparameters, thereby constructing a high-precision pavement condition index (PCI) decay prediction framework. Using measured time-series data from two runways in China, covering 6 types of key indicators such as service time, rainfall, and traffic load, the prediction accuracy and generalization ability of different models were calculated and analyzed, and SHAP and a generalized additive model were introduced to analyze the nonlinear influence mechanisms of key features. The research results indicate that compared to the traditional LSTM and single improved models, the CNN-xLSTM-Attention model has the optimal comprehensive performance, effectively solving the long-sequence gradient problem and the feature coupling challenge; the coefficient of determination (R2) of the test set reaches 0.923; the mean absolute percentage error (MAPE) is only 1.778%. The explanatory analysis reveals that service time (contribution rate of 44.8%) and annual average rainfall (contribution rate of 33.6%) are the primary factors dominating the decay of PCI; traffic load and temperature indicators show a significant threshold effect; increasing the surface course thickness has a significant inhibitory effect on delaying the decay. It is thus evident that the established pavement performance prediction model and its explanatory analysis method can accurately capture the temporal evolution patterns and key influence mechanisms of PCI, providing reliable data support for the phased scientific maintenance and operational decisions of airport runways.
-
图 6 基于泰勒图的模型测试集性能对比
Figure 6. Performance comparison of models on the test set based on Taylor diagram
图 7 不同模型预测值与真实值对比
Figure 7. Comparison of predicted values and true values by different models
图 8 基于SHAP模型的R1道面PCI衰减值特征重要性分析
Figure 8. Feature importance analysis of R1 pavement PCI deterioration value based on SHAP model
表 1 跑道道面区块和单元划分结果
Table 1. Division results of runway pavement blocks and units
跑道 单元总数/个 区块 单元数量/个 R1 240 R1-01 40 R1-02 53 R1-03 54 R1-04 53 R1-05 40 R2 400 R2-01 66 R2-02 89 R2-03 90 R2-04 89 R2-05 66 
下载: 导出CSV
表 2 道面性能影响因素类型
Table 2. Types of factors influencing pavement performance
变量类型 描述 均值 范围 环境参数 x2/mm 年平均降雨量 765.2 [200，1 798] x4/d 年气温超过30 ℃的天数 35.3 [11，49] x5/d 年气温低于-5 ℃的天数 22.5 [0, 45] 荷载参数 x3/万架 经折算后的年飞机运行架次 14.9 [1.2，26.8] 路龄 x1/年 距离建成或上一次大中修的时间 4 [1，7] 结构参数 x6/mm 面层厚度 160 [100，220]
下载: 导出CSV
表 3 相关超参数优化结果
Table 3. Optimization results of related hyperparameters
模块 超参数 范围 最优值 卷积层 C {16，32，64，128} 128 xLSTM 每个分支（sLSTM、mLSTM）的隐藏单元数ℎ {32，64，128} 128 Dropout比例p （0，0.4） 0.249 Attention 注意力全连接层维度datt {8，16，32，64} 64 训练相关超参数 学习率η （1.0×10-4，1.0×10-2） 2.2×10-3 Batch size {8，16，32，64} 8
下载: 导出CSV
-
[1] 凌建明, 方意心, 张家科, 等. 机场智能跑道体系架构与关键技术[J]. 土木工程学报, 2022, 55(2): 120-128.LING Jian-ming, FANG Yi-xin, ZHANG Jia-ke, et al. Framework and key technologies of airport smart runway[J]. China Civil Engineering Journal, 2022, 55(2): 120-128. [2] HOU T X, LIU S F, MAO W Y, et al. Stochastic vibration analysis of aircraft-rigid pavement system under random aircraft parameters[J]. International Journal of Pavement Engineering, 2026, 27(1): 2666268. doi: 10.1080/10298436.2026.2666268 [3] RAGUSA E, COSTA A, DI GRAZIANO A. Exploring new computational strategies for managing maintenance activities of airport pavement systems[J]. International Journal of Pavement Research and Technology, 2024, 17(1): 80-101. doi: 10.1007/s42947-022-00222-6 [4] 史保华, 许巍, 齐勇, 等. 机场沥青道面管理与预防性维护对策[J]. 交通运输工程学报, 2008, 8(2): 74-79.SHI Bao-hua, XU Wei, QI Yong, et al. Management and preventive maintenance countermeasures of airport asphalt pavement[J]. Journal of Traffic and Transportation Engineering, 2008, 8(2): 74-79. [5] SONG Y, WANG Y D, HU X, et al. An efficient and explainable ensemble learning model for asphalt pavement condition prediction based on LTPP dataset[J]. IEEE Transactions on Intelligent Transportation Systems, 2022, 23(11): 22084-22093. doi: 10.1109/TITS.2022.3164596 [6] SHOLEVAR N, GOLROO A, ESFAHANI S R. Machine learning techniques for pavement condition evaluation[J]. Automation in Construction, 2022, 136: 104190. doi: 10.1016/j.autcon.2022.104190 [7] LI J L, YIN G H, WANG X F, et al. Automated decision making in highway pavement preventive maintenance based on deep learning[J]. Automation in Construction, 2022, 135: 104111. doi: 10.1016/j.autcon.2021.104111 [8] SUN X, WANG H L, MEI S L. Explainable highway performance degradation prediction model based on LSTM[J]. Advanced Engineering Informatics, 2024, 61: 102539. doi: 10.1016/j.aei.2024.102539 [9] CHEN S T, CAO J D, WAN Y, et al. A novel CPO-CNN-LSTM based deep learning approach for multi-time scale deflection basin area prediction in asphalt pavement[J]. Construction and Building Materials, 2025, 458: 139540. doi: 10.1016/j.conbuildmat.2024.139540 [10] DENG Y, SHI X M. Short-term predictions of asphalt pavement rutting using deep-learning models[J]. Journal of Transportation Engineering, Part B: Pavements, 2024, 150(2): 04024004. doi: 10.1061/JPEODX.PVENG-1409 [11] WU Y N, ZHANG Q F, WANG Y, et al. Advanced hybrid CNN-GRU model for IRI prediction in flexible asphalt pavements[J]. Journal of Transportation Engineering, Part B: Pavements, 2025, 151(2): 04025003. doi: 10.1061/JPEODX.PVENG-1570 [12] LI J L, ZHANG Z S, WANG X F, et al. Intelligent decision-making model in preventive maintenance of asphalt pavement based on PSO-GRU neural network[J]. Advanced Engineering Informatics, 2022, 51: 101525. doi: 10.1016/j.aei.2022.101525 [13] BECK M, PÖPPEL K, SPANRING M, et al. xLSTM: Extended long short-term memory[J]. Advances in Neural Information Processing Systems, 2024, 37: 107547-107603. [14] HAN C J, MA T, GU L H, et al. Asphalt pavement health prediction based on improved transformer network[J]. IEEE Transactions on Intelligent Transportation Systems, 2023, 24(4): 4482-4493. doi: 10.1109/TITS.2022.3229326 [15] 张兵, 周丹丹, 孙健, 等. 基于双向长短期记忆网络的公交到站时间预测模型[J]. 交通运输系统工程与信息, 2023, 23(2): 148-160.ZHANG Bing, ZHOU Dan-dan, SUN Jian, et al. Bus arrival time prediction model based on bidirectional long short-term memory network[J]. Journal of Transportation Systems Engineering and Information Technology, 2023, 23(2): 148-160. [16] OYEWOLA D O, AKINWUNMI S A, OMOTEHINWA T O. Deep LSTM and LSTM-Attention Q-learning based reinforcement learning in oil and gas sector prediction[J]. Knowledge-based Systems, 2024, 284: 111290. doi: 10.1016/j.knosys.2023.111290 [17] MH/T 5024—2019, 民用机场道面评价管理技术规范[S].MH/T 5024—2019, Technical Specification for Airport Pavement Evaluation and Management[S]. [18] LI Y, ZOU Z B, ZHANG J P, et al. Study on the evolution of airport asphalt pavement integrated distress based on association rule mining[J]. Construction and Building Materials, 2023, 369: 130565. doi: 10.1016/j.conbuildmat.2023.130565 [19] 李龙海, 伍思铭. 机场刚性道面等级评价与参数敏感性分析[J]. 重庆交通大学学报(自然科学版), 2026, 45(2): 14-23.LI Long-hai, WU Si-ming. Evaluation of airport rigid pavement grades and parameter sensitivity[J]. Journal of Chongqing Jiaotong University (Natural Sciences), 2026, 45(2): 14-23. [20] BHANDARI S, LUO X H, WANG F. Understanding the effects of structural factors and traffic loading on flexible pavement performance[J]. International Journal of Transportation Science and Technology, 2023, 12(1): 258-272. doi: 10.1016/j.ijtst.2022.02.004 [21] KIRANYAZ S, AVCI O, ABDELJABER O, et al. 1D Convolutional neural networks and applications: A survey[J]. Mechanical Systems and Signal Processing, 2021, 151: 107398. doi: 10.1016/j.ymssp.2020.107398 [22] 张颖超, 安然, 陈昕, 等. 基于多时距风速波动过程划分的高铁沿线风速预测[J]. 交通运输工程学报, 2025, 25(3): 362-379.ZHANG Ying-chao, AN Ran, CHEN Xin, et al. Wind speed prediction along high-speed railway based on multi-time-interval wind speed fluctuation process division[J]. Journal of Traffic and Transportation Engineering, 2025, 25(3): 362-379. [23] 王晓明, 汪帆, 翟岸, 等. 退化RC桥梁检测-维护规划的贝叶斯决策方法[J]. 交通运输工程学报, 2025, 25(3): 130-143. doi: 10.19818/j.cnki.1671-1637.2025.03.008WANG Xiao-ming, WANG Fan, ZHAI An, et al. Bayesian decision method of inspection and maintenance planning for deteriorating RC bridges[J]. Journal of Traffic and Transportation Engineering, 2025, 25(3): 130-143. doi: 10.19818/j.cnki.1671-1637.2025.03.008 [24] HE Y Z, GAO S J, LI Y, et al. Adaptive machine learning framework: Predicting UHPC performance from data to modelling[J]. Results in Engineering, 2025, 27: 106724. doi: 10.1016/j.rineng.2025.106724 [25] LI Z Q. Extracting spatial effects from machine learning model using local interpretation method: An example of SHAP and XGBoost[J]. Computers, Environment and Urban Systems, 2022, 96: 101845. doi: 10.1016/j.compenvurbsys.2022.101845 [26] 何印章, 苏浩然, 郝宁远, 等. 基于机器学习和SHAP模型的超高性能混凝土性能预测与分析框架[J]. 中国公路学报, 2026, 39(4): 261-282.HE Yin-zhang, SU Hao-ran, HAO Ning-yuan, et al. A predictive and analytical framework for ultra-high performance concrete properties using machine learning and SHAP modeling[J]. China Journal of Highway and Transport, 2026, 39(4): 261-282. [27] YANG X, ZHANG J Q, LIU W B, et al. Automation in road distress detection, diagnosis and treatment[J]. Journal of Road Engineering, 2024, 4(1): 1-26. doi: 10.1016/j.jreng.2024.01.005 [28] PINHEIRO M G, MADEIRA S C, FRANCISCO A P. Short-term electricity load forecasting: A systematic approach from system level to secondary substations[J]. Applied Energy, 2023, 332: 120493. doi: 10.1016/j.apenergy.2022.120493 -
图(9) / 表(3)
计量
- 文章访问数: 30
- HTML全文浏览量: 7
- PDF下载量: 3
- 被引次数: 0
