Road Weather Condition Prediction Based on Numerical Weather Prediction and Machine Learning Technology
-
摘要: 道路气象状况与交通安全密切相关,路面湿滑、结冰易引发行车事故,因此需要实现准确且及时的道路气象状况预测。利用雅康高速路段内3个地面观测点的道路气象状况观测数据,以及对应区域的连续24 h数值天气预报数据,构建决策树模型,建立数值天气预报结果与多种道路气象状况类别之间的对应关系,实现未来连续24 h的道路气象状况预测。结果表明,针对3个地面观测点的5类道路气象状况,在交叉验证实验中,所提出模型的平均准确率为89.79%,在外推实验中,模型对未来第6 h预测的平均准确率为64.73%,未来第12 h预测为77.30%,未来第18 h预测为80.19%,未来第24 h预测为70.41%。研究方法可以有效实现连续空间覆盖、长时间的道路气象状况预测,为交通运输安全、公众出行决策、气象预报服务等方面提供重要参考信息。Abstract: Road weather conditions are closely related to traffic safety, as slippery and icy roads can easily lead to accidents. Therefore, accurate and timely predictions of road weather conditions are essential. The data used in the present study included the observation data of road weather conditions from three ground observation stations along the Yakang highway and the 24-hour numerical weather prediction data for the corresponding area. Based on a decision tree model, the corresponding relationship between numerical weather prediction results and various types of road weather conditions was established, enabling predictions of road weather conditions for the next 24 hours. The results show that, for the five types of road weather conditions at three ground observation stations, the average accuracy of cross-validation for our proposed model was 89.79%. In the extrapolation experiment, the average accuracy of prediction for the next 6 hours was 64.73%, for the next 12 hours was 77.30%, for the next 18 hours was 80.19%, and for the next 24 hours was 70.41%. Our research method effectively achieved continuous spatial coverage and long-term prediction of road weather conditions, providing important reference information for traffic safety, public travel decision-making, and weather forecasting services.
-
Key words:
- road weather condition /
- prediction model /
- decision tree /
- machine learning /
- meteorological service
-
表 1 数值天气预报数据示例
时间 风速/(m·s-1) 风向/° 气压/Pa … 温度/℃ 湿度/% 2022/12/16 13:00 1.80 255.33 88 869.75 … 0.44 71.2 2022/12/16 14:00 1.30 261.52 88 970.95 … 0.85 69.9 2022/12/16 15:00 0.50 263.28 88 976.03 … 0.98 73.3 2022/12/16 16:00 0.76 359.83 89 072.39 … 0.88 66.7 2022/12/16 17:00 2.00 10.68 89 248.05 … 0.24 60.7 表 2 道路气象状况观测数据示例
时间 喇叭河隧道出口 天河隧道出口 李子坪大桥 气象条件 标记码 气象条件 标记码 气象条件 标记码 2022/12/16 13:00 干燥 0 潮 1 冰 4 2022/12/16 14:00 干燥 0 湿 2 冰水混合物 3 2022/12/16 15:00 湿 2 湿 2 湿 2 2022/12/16 16:00 潮 1 湿 2 湿 2 2022/12/16 17:00 潮 1 湿 2 湿 2 表 3 模型超参数的搜索范围
超参数 中文释义 搜索范围 搜索间隔 n_estimators 学习器的数量 25~300 25 eta 整体学习速率 0.1~0.5 0.1 max_depth 树的最大深度 1~7 1 min_child_weight 最小叶子节点权重 1~7 1 subsample 随机采样比例 0.7~1 0.1 colsample_bytree 随机采样列数比例 0.7~1 0.1 表 4 模型性能对比
模型 喇叭河隧道出口 天河隧道出口 李子坪大桥 Train_cost/s Test_cost/s SVC 74.22% 59.30% 52.72% 0.025 0.022 KNN 78.92% 69.57% 65.40% - 0.119 RF 91.11% 86.43% 85.33% 1.250 0.010 RF 91.11% 86.43% 85.33% 1.250 0.010 XGBoost 93.55% 88.87% 86.96% 0.168 0.003 表 5 不同预测时间段外推性能(单位:%)
预测时间 喇叭河隧道出口 天河隧道出口 李子坪大桥 平均 24小时整体 65.68 75.83 66.67 69.39 第1小时 45.83 70.83 56.52 57.73 第6小时 58.33 75.00 60.87 64.73 第12小时 79.17 87.50 65.22 77.30 第18小时 83.33 83.33 73.91 80.19 第24小时 62.50 79.17 69.57 70.41 表 6 不同气象状况类别外推性能(单位:%)
气象状况类别 喇叭河隧道出口 天河隧道出口 李子坪大桥 平均 干燥 76.69 83.64 80.17 80.17 潮 69.35 71.72 71.85 70.97 湿 60.87 71.43 67.92 66.74 冰水混合物 55.56 - 60.00 57.78 冰 - - 50.00 50.00 4类事故易发类别 66.97 71.58 69.61 69.39 注:“-”表示该观测点无此气象状况类别数据。 -
[1] 沈萍月, 李建, 潘娅英, 等. 高速公路气象服务技术及应用[J]. 气象科技, 2014, 42(6): 1 159-1 162. [2] 康延臻, 王式功, 杨旭, 等. 高速公路交通气象监测预报服务研究进展[J]. 干旱气象, 2016, 34(4): 591-603. [3] 黄树燕, 史彩霞, 谭天信, 等. 广西主要高速公路气象灾害风险调查分析[J]. 气象研究与应用, 2017, 38(3): 99-104. [4] 欧彦, 蒲翔, 周旭驰, 等. 路面结冰监测技术研究进展[J]. 公路, 2013 (4): 191-196. [5] NORRMAN J. Slipperiness on roads-an expert system classification[J]. Meteorological applications, 2000, 7(1): 27-36. [6] SASS B H. A numerical model for prediction of road temperature and ice[J]. Journal of Applied Meteorology and Climatology, 1992, 31 (12): 1 499-1 506. [7] SHAO J, LISTER P J. An automated nowcasting model of road surface temperature and state for winter road maintenance[J]. Journal of Applied Meteorology and Climatology, 1996, 35(8): 1 352-1 361. [8] 张宏芳, 卢珊, 沈姣姣, 等. 陕西道路结冰时空变化特征及其风险预警模型[J]. 干旱气象, 2020, 38(5): 878-885. [9] 张兴山, 王娟怀, 叶培龙, 等. 基于统计分析的城市道路结冰预报模型研究[J]. 现代农业研究, 2022, 28(11): 110-115. [10] 李刚, 罗喜平, 韩晓令. 贵州中西部道路结冰气象影响因子检验评估[J]. 贵州气象, 2016, 40(4): 6-10. [11] 孟思彤, 白爱娟, 郑自君, 等. 川西雅康高速道路冬季路面结冰预警阈值分析[J]. 成都信息工程大学学报, 2023, 38(5): 615-620. [12] 李细生, 张华, 喻雨知, 等. 基于机器学习分类算法的臭氧浓度等级预报在长沙的应用[J]. 热带气象学报, 2023, 39(4): 453-461. [13] ZHAO B, YU L, WANG C, et al. Urban air pollution mapping using fleet vehicles as mobile monitors and machine learning[J]. Environmental Science & Technology, 2021, 55(8): 5 579-5 588. [14] GUO R, QI Y, ZHAO B, et al. High-resolution urban air quality mapping for multiple pollutants based on dense monitoring data and machine learning[J]. International Journal of Environmental Research and Public Health, 2022, 19(13): 8 005. [15] 康俊锋, 冯松江, 邹倩, 等. 基于机器学习的非线性局部Lyapunov向量集合预报订正[J]. 物理学报, 2022, 71(8): 46-54 [16] 杨鹏飞, 孙先波. 基于支持向量机算法的道路结冰预测模型研究[J]. 湖北民族大学学报(自然科学版), 2020, 38(3): 355-360. [17] 陈凯, 徐国庆, 熊焘, 等. 基于C4.5决策树算法的道路结冰预报模型构建与应用[J]. 交通运输研究, 2018, 4(1): 57-62. [18] 鲍丽丽, 程鹏, 王小勇, 等. 基于Logistic回归和神经网络的甘肃省道路结冰预警模型研究[J]. 干旱气象, 2024, 42(1): 137-145. [19] KIM S, LEE J, YOON T. Road surface conditions forecasting in rainy weather using artificial neural networks[J]. Safety science, 2021, 140: 105302. [20] 沈学顺, 王建捷, 李泽椿, 等. 中国数值天气预报的自主创新发展[J]. 气象学报, 2020, 78(3): 451-476. [21] BENJAMIN S G, BROWN J M, BRUNET G, et al. 100 years of progress in forecasting and NWP applications[J]. Meteor Monogr, 59(1): 13.1-13.67. [22] 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. 2016: 785-794. [23] WONG P Y, LEE H Y, CHEN Y C, et al. Using a land use regression model with machine learning to estimate ground level PM2.5[J]. Environmental Pollution, 2021, 277: 116846. [24] FENG S, TAN Y, KANG J, et al. Bias correction of tropical cyclone intensity for ensemble forecasts using the XGBoost method[J]. Weather and Forecasting, 2024, 39(2): 323-332. -