Prediction of potential risk paths in natural disaster network along railway lines based on text mining
-
摘要:
为揭示铁路沿线自然灾害链式传播机理，量化其对铁路运输安全的威胁并解析风险演化规律，构建了基于文本挖掘与复杂网络的铁路沿线自然灾害风险路径预测模型。以铁路沿线历史灾害文本数据集为基础，改进了文本挖掘技术，优化词袋模型并结合关键字提取方法，精准抽取了灾害数据中的铁路站点、致灾因子、后果及严重程度等核心要素；基于复杂网络理论，以关键词为节点(3类实体)，以同事件共现语义关联为边(6类关系)，建立了铁路灾害风险异构网络；设计多元路径搜索算法遍历网络拓扑，融合关联共现矩阵量化节点间风险传导强度，实现了多类型传播路径的过程化挖掘。分析结果表明：模型接收者操作特征曲线接近左上角，曲线下面积为0.938，准确率达94.873%，F1分数为0.899；量化输出节点对间风险传导值，成功定位高概率灾害链为芒康站→强降雪→香格里拉站→人员受灾→电力设备损毁(风险值0.866)、江达站→贡觉站→强对流天气→人员伤亡(风险值0.841)。所得模型通过文本驱动构建灾害语义网络，可实现铁路风险路径的定量化预测与关键传导链识别，精准定位二次灾害高发链路，为复杂环境铁路的风险主动防控提供支持。
Abstract:To reveal the chain propagation mechanism of natural disasters along railway lines, quantify their threats to railway transportation safety, and analyze the risk evolution laws, a prediction model for natural disaster risk paths along railway lines based on text mining and complex networks was constructed. Based on the historical disaster text dataset along railway lines, the text mining technology was improved; the bag-of-words model was optimized, and combined with the keyword extraction method, core elements such as railway stations, disaster-causing factors, consequences, and severity levels in the disaster data were accurately extracted; based on complex network theory, a heterogeneous network of railway disaster risks was established, with keywords as nodes (3 types of entities) and co-occurrence semantic associations in the same event as edges (6 types of relationships); a multi-path search algorithm was designed to traverse the network topology, and the associated co-occurrence matrix was integrated to quantify the risk transmission intensity between nodes, realizing the procedural mining of multiple types of propagation paths. The analysis results show that the receiver operating characteristic curve of the model is close to the upper left corner; the area under the curve is 0.938; the accuracy reaches 94.873%, and the F1 score is 0.899; the risk transmission values between node pairs are quantitatively output, and high-probability disaster chains are successfully located as: Markam Station → heavy snowfall → Shangri-La Station → personnel disaster → power equipment damage (risk value 0.866), and Jomda Station → Gonjo Station → severe convective weather → casualties (risk value 0.841). The obtained model constructs a disaster semantic network driven by text, which can achieve the quantitative prediction of railway risk paths and the identification of key transmission chains, accurately locate the high-incidence links of secondary disasters, and provide support for the proactive risk prevention and control of railways in complex environments.
-
图 2 高原铁路沿线自然灾害异构网络
Figure 2. Heterogeneous network of natural disasters along the plateau railway
图 3 不同方法在同质路径中的ROC曲线比较
Figure 3. ROC curve comparison for different methods in homogeneous paths
图 4 不同方法在异质路径中的ROC曲线
Figure 4. ROC curve comparison for different methods in heterogeneous paths
图 5 不同方法在所有混合路径中的ROC曲线比较
Figure 5. ROC curve comparison for different methods in all mixed paths
图 6 PRP模型与其他对比方法t检验的箱线图与数据点叠加图
Figure 6. Box plot and data point overlay for t-test of PRP model and other comparative methods
图 7 PRP模型对高原铁路沿线自然灾害风险路径预测结果的可视化
Figure 7. Visualization of PRP model for predicting natural disaster risk paths along the plateau railway
表 1 典型链路预测方法
Table 1. Typical link prediction methods
方法 分数计算公式 启发式方法 局部算法 共同邻居(Common Neighbors, CN) $S_{i j}=|\varGamma(i) \cap \varGamma(j)|$ 优先依附(Preferential Attachment, PA) Sij=kikj Jaccard系数(JC) $S_{i j}=\frac{|\varGamma(i) \bigcap \varGamma(j)|}{|\varGamma(i) \bigcup \varGamma(j)|}$ 全局算法 平均通勤时间(Average Commute Time, ACT) $S_{i j}=\frac{1}{l_{i i}^{+}+l_{j j-}^{+}+2 l_{i j}^{+}}$ 矩阵森林(Matrix Forest, MF) Sij=lij 图嵌入方法 随机游走(Random Walk, RW) Sij=qij+qji 节点嵌入算法(Node2Vec) 图神经网络方法 图采样聚合网络(Graph Sample and Aggregate, GraphSAGE) 图卷积神经网络(Graph Convolutional Network, GCN) 
下载: 导出CSV
表 2 高原铁路沿线部分自然灾害原始数据示例
Table 2. Example of original data on some natural disasters along the plateau railway
事件ID 事件类型 公告日期 影响地区 灾情概述(节选) HL2014000035 洪涝 2014-07-22 四川省绵阳、德阳、乐山 7月19日以来，部分地区出现大雨、暴雨过程，引发洪涝灾害，截至7月22日统计，造成绵阳、德阳、乐山3市6个县(市、区)13万人受灾，100余间房屋倒塌，700余间损坏；直接经济损失6 800余万元 DF2014000031 大风 2014-07-27 西藏自治区拉萨市林周县、昌都地区察雅县、阿里地区措勤县 7月25日，拉萨市林周县、昌都地区察雅县、阿里地区措勤县遭受风雹灾害，2人因雷击死亡 DZA201400004 地震 2014-08-05 四川省凉山彝族自治州宁南县、金阳县 8月3日16:30分，云南省昭通市鲁甸县发生6.5级地震，震源深度12 km。截至8月5日初步统计，地震造成四川省凉山彝族自治州宁南县、金阳县1.16万人受灾，近500余间房屋严重损坏，2 000余间一般损坏 NSL201400004 泥石流 2014-08-09 西藏自治区昌都县卡若镇 8月7日，昌都地区昌都县卡若镇发生泥石流灾害，造成1人死亡
下载: 导出CSV
表 3 每种链接类型的样本数量
Table 3. Number of samples for each link type
路径类型 P-P C-C S-S P-C P-S C-S 整体 样本数量 621 44 118 394 840 159 2 176
下载: 导出CSV
表 4 不同方法在不同路径类型中的AUC与准确率
Table 4. AUC and accuracy of different methods in different path types
方法 指标 P-P C-C S-S P-C P-S C-S 混合路径 CN AUC 0.858 0.886 0.927 0.901 0.937 0.904 0.928 准确率/% 94.502 90.323 89.474 91.223 88.688 90.226 93.879 F1分数 0.873 0.875 0.874 0.841 0.870 0.879 0.892 PA AUC 0.829 0.867 0.859 0.845 0.936 0.904 0.906 准确率/% 97.608 87.097 91.729 89.096 89.465 86.466 93.692 F1分数 0.837 0.843 0.849 0.820 0.833 0.871 0.880 JC AUC 0.870 0.780 0.903 0.894 0.850 0.906 0.839 准确率/% 97.032 90.323 90.977 91.755 91.144 87.218 94.397 F1分数 0.860 0.864 0.875 0.863 0.825 0.854 0.884 ACT AUC 0.835 0.869 0.866 0.839 0.891 0.899 0.859 准确率/% 97.493 90.323 87.218 92.420 89.399 87.218 93.317 F1分数 0.852 0.845 0.865 0.880 0.888 0.865 0.890 MF AUC 0.823 0.853 0.833 0.859 0.835 0.835 0.792 准确率/% 97.562 83.871 80.451 88.697 88.623 88.948 93.692 F1分数 0.869 0.834 0.870 0.849 0.853 0.878 0.881 RW AUC 0.866 0.881 0.865 0.851 0.930 0.854 0.891 准确率/% 94.430 87.097 85.714 88.963 90.109 86.466 93.042 F1分数 0.891 0.853 0.893 0.892 0.850 0.889 0.886 Node2Vec AUC 0.894 0.893 0.923 0.887 0.900 0.904 0.859 准确率/% 97.125 90.323 87.970 88.564 90.045 88.722 93.821 F1分数 0.885 0.891 0.880 0.874 0.900 0.880 0.896 GraphSAGE AUC 0.830 0.848 0.881 0.826 0.887 0.842 0.843 准确率/% 96.940 87.419 82.707 87.766 88.559 90.451 93.184 F1分数 0.892 0.866 0.866 0.882 0.882 0.874 0.894 GCN AUC 0.898 0.875 0.929 0.888 0.905 0.915 0.875 准确率/% 97.976 93.545 90.226 91.755 90.304 87.970 94.196 F1分数 0.861 0.863 0.889 0.884 0.901 0.891 0.891 PRP AUC 0.909 0.918 0.942 0.928 0.968 0.943 0.938 准确率/% 97.884 93.548 89.474 90.957 89.528 90.977 94.873 F1分数 0.896 0.889 0.885 0.897 0.899 0.897 0.899
下载: 导出CSV
表 5 PRP模型与其他对比方法的t检验结果
Table 5. The t-test results of PRP model compared with other methods
对比方法 均值差 t统计量 P值 95%置信区间 结论 PRP-CN 0.013 7 25.973 7 ＜0.001 [0.012 7, 0.014 8] 显著不同 PRP-PA 0.008 0 14.844 4 ＜0.001 [0.007 0, 0.009 1] 显著不同 PRP-JC 0.078 6 52.807 8 ＜0.001 [0.075 7, 0.081 5] 显著不同 PRP-ACT 0.038 0 55.874 2 ＜0.001 [0.036 7, 0.039 4] 显著不同 PRP-MF 0.063 0 129.819 1 ＜0.001 [0.062 0, 0.063 9] 显著不同 PRP-RW 0.233 1 67.683 0 ＜0.001 [0.226 3, 0.239 8] 显著不同 PRP-Node2Vec -0.004 4 -14.200 8 ＜0.001 [-0.005 0, -0.003 8] 显著不同 PRP-GraphSAGE 0.250 3 95.019 1 ＜0.001 [0.245 1, 0.255 5] 显著不同 PRP-GCN 0.115 7 62.523 8 ＜0.001 [0.112 1, 0.119 3] 显著不同
下载: 导出CSV
表 6 不同路径类型的预测结果
Table 6. Prediction results of different path types
路径类型 预测结果 分数 P-P 贡觉-芒康 0.917 江达-贡觉 0.911 昌都-甘孜 0.910 C-C 降雪-泥石流 0.951 山洪-滑坡 0.940 风雹-山体崩塌 0.931 S-S 人员受灾-电力设备损毁 0.945 房屋倒损-电力设备损毁 0.906 房屋倒损-供水设备损毁 0.902 P-C 甘孜-风雹 0.918 雅安-地震 0.916 眉山-洪涝 0.911 P-S 自贡-经济损失 0.955 香格里拉-人员受灾 0.953 洛隆-房屋倒损 0.951 C-S 强对流天气-人员伤亡 0.929 大风-道路损毁 0.913 滑坡-桥梁损毁 0.908
下载: 导出CSV
表 7 风险传导过程量化分析
Table 7. Quantitative analysis of risk transmission process
路径 传导步骤 当前节点(风险值) 预测分数 传导至下一节点风险值 结论 1 起点 P1(SP1=0.6) 风险放大：灾害后果严重 第1跳 P1(SP1=0.6) SP1C1=0.901 0.6×0.901=0.541 第2跳 C1(SC1=0.541) SC1P2=0.890 0.541×0.890=0.481 第3跳 P2(SP2=0.481) SP2S1=0.953 0.481×0.953=0.458 第4跳 S1(SS1=0.458) SS1S2=0.945 0.458×0.945=0.433 后果与放大 S2(SS2=0.433) wS2=2.0 0.433×2.0=0.866 2 起点 P3(SP3=0.7) 风险放大：灾害后果严重 第1跳 P3(SP3=0.7) SP3P4=0.911 0.7×0.911=0.638 第2跳 P4(SP4=0.638) SP4C2=0.946 0.638×0.946=0.603 第3跳 C2(SC2=0.603) SC2S3=0.929 0.603×0.929=0.560 后果与放大 S3(SS3=0.560) wS3=1.5 0.560×1.5=0.841
下载: 导出CSV
-
[1] LU C F, CAI C X. Challenges and countermeasures for construction safety during the Sichuan-Tibet Railway project[J]. Engineering, 2019, 5: 833-838. doi: 10.1016/j.eng.2019.06.007 [2] 李远远, 梅红波, 任晓杰, 等. 基于确定性系数和支持向量机的地质灾害易发性评价[J]. 地球信息科学学报, 2018, 20(12): 1699-1709.LI Yuan-yuan, MEI Hong-bo, REN Xiao-jie, et al. Geological disaster susceptibility evaluation based on certainty factor and support vector machine[J]. Geo-Information Science, 2018, 20(12): 1699-1709. [3] 夏溪蔓, 孟学雷, 程晓卿, 等. 基于内外双视角的高速铁路风险预测[J]. 铁道科学与工程学报, 2025, 22(7): 2921-2931.XIA Xi-man, MENG Xue-lei, CHENG Xiao-qing, et al. High-speed railway risk prediction based on internal and external perspectives[J]. Journal of Railway Science and Engineering, 2025, 22(7): 2921-2931. [4] 王浙涛, 陈如海, 魏国俊. 地质灾害危险性评估在铁路建设中的重要性[J]. 甘肃科学学报, 2003, 15: 72-75.WANG Zhe-tao, CHEN Ru-hai, WEI Guo-jun. Significance of risk assessment of geological hazards in railway construction[J]. Journal of Gansu Sciences, 2003, 15: 72-75. [5] LAN T W, FAN C J, LI S, et al. Probabilistic prediction of mine dynamic disaster risk based on multiple factor pattern recognition[J]. Advances in Civil Engineering, 2018, 2018: 7813931. doi: 10.1155/2018/7813931 [6] 宋哲, 郭湛, 习年生, 等. 基于概率分析的铁路安全风险传递网络模型研究[J]. 铁道运输与经济, 2024, 46(4): 142-152.SONG Zhe, GUO Zhan, XI Nian-sheng, et al. Research on railway safety risk transfer network model based on probability analysis[J]. Railway Transport and Economy, 2024, 46(4): 142-152. [7] 吴晶晶, 江思义, 吴秋菊, 等. 基于GIS与BP神经网络的崩塌滑坡地质灾害易发性预测[J]. 资源信息与工程, 2021, 36(4): 100-107.WU Jing-jing, JIANG Si-yi, WU Qiu-ju, et al. Landslide geological hazard vulnerability evaluation based on GIS and BP neural network[J]. Resource Information and Engineering, 2021, 36(4): 100-107. [8] WANG X D, ZHANG C B, WANG C, et al. GIS-based for prediction and prevention of environmental geological disaster susceptibility: From a perspective of sustainable development[J]. Ecotoxicology and Environmental Safety, 2021, 226: 112881. doi: 10.1016/j.ecoenv.2021.112881 [9] 符洪恩, 高艺桔, 冯莹莹, 等. 基于GA-SVR-C的城市暴雨洪涝灾害危险性预测——以深圳市为例[J]. 人民长江, 2021, 52(8): 16-21.FU Hong-en, GAO Yi-ju, FENG Ying-ying, et al. Hazard prediction of urban rainstorm and flood disasters based on GA-SVR-C model: Case study of Shenzhen City[J]. Yangtze River, 2021, 52(8): 16-21. [10] 宋国策, 王高磊, 卢大玮, 等. 基于ISM-ANP的铁路外部环境安全隐患风险评价模型及应用[J]. 铁道运输与经济, 2024, 46(6): 161-168.SONG Guo-ce, WANG Gao-lei, LU Da-wei, et al. Risk evaluation model of safety hazards in railway external environment based on ISM-ANP and its application[J]. Railway Transport and Economy, 2024, 46(6): 161-168. [11] 龚艳冰, 向林, 刘高峰. 基于高斯过程回归模型的洪涝灾害损失预测研究——以重庆市为例[J]. 长江流域资源与环境, 2019, 28(6): 1502-1510.GONG Yan-bing, XIANG Lin, LIU Gao-feng. Study on flood disaster loss prediction based on gaussian process regression model: A case study of Chongqing city[J]. Resources and Environment in the Yangtze Basin, 2019, 28(6): 1502-1510. [12] CAO Y, YIN K D, LI X M. Prediction of direct economic loss caused by marine disasters based on the improved GM (1, 1) model[J]. Journal of Grey System, 2020, 32(1): 133-145. [13] 李博, 冯俏彬, 戚克维. 基于改进的神经网络模型预测气象灾害经济损失——以广东省台风灾害为例[J]. 重庆理工大学学报(自然科学), 2021, 35(4): 247-253.LI Bo, FENG Qiao-bin, QI Ke-wei. Economic loss prediction of meteorological disaster based on improved neural network model: Taking the Guangdong typhoon as an example[J]. Journal of Chongqing University of Technology(Natural Science), 2021, 35(4): 247-253. [14] ZHANG Y, HAO Y H. Loss prediction of mountain flood disaster in villages and towns based on rough set RBF neural network[J]. Neural Computing and Applications, 2022, 34(5): 2513-2524. [15] HUANG X, SONG J Y, JIN H D. The casualty prediction of earthquake disaster based on extreme learning machine method[J]. Natural Hazards, 2020, 102: 873-886. doi: 10.1007/s11069-020-03937-6 [16] 胡忠君, 刘艳秋, 李佳. 基于改进GM(1, 1)的洪涝灾害应急物资动态需求预测[J]. 系统仿真学报, 2019, 31(4): 702-709.HU Zhong-jun, LIU Yan-qiu, LI Jia. Dynamic demand forecast of emergency materials for flood disasters based on improved GM(1, 1) model[J]. Journal of System Simulation, 2019, 31(4): 702-709. [17] 刘芳, 冯丹, 宫雪然. 基于IACO-BP算法的洪涝灾害应急物资需求预测[J]. 沈阳工业大学学报, 2019, 41(3): 332-338.LIU Fang, FENG Dan, GONG Xue-ran. Demand predicting of emergency supplies for flood disaster based on IACO-BP algorithm[J]. Journal of Shenyang University of Technology, 2019, 41(3): 332-338. [18] HOU L, WU X G, WU Z, et al. Pattern identification and risk prediction of domino effect based on data mining methods for accidents occurred in the tank farm[J]. Reliability Engineering and System Safety, 2020, 193: 106646. doi: 10.1016/j.ress.2019.106646 [19] 许佳, 丁超, 张昕禹, 等. 基于复杂网络的城市内涝灾害应急管理研究综述[J]. 自然灾害学报, 2024, 33(3): 1-16.XU Jia, DING Chao, ZHANG Xin-yu, et al. Review of complex network-based urban waterlogging disaster emergency management research[J]. Journal of Natural Disasters, 2024, 33(3): 1-16. [20] 马超群, 张爽, 陈权, 等. 客流特征视角下的轨道交通网络特征及其脆弱性[J]. 交通运输工程学报, 2020, 20(5): 208-216. doi: 10.19818/j.cnki.1671-1637.2020.05.017MA Chao-qun, ZHANG Shuang, CHEN Quan, et al. Characteristics and vulnerability of rail transit network based on perspective of passenger flow characteristic[J]. Journal of Traffic and Transportation Engineering, 2020, 20(5): 208-216. doi: 10.19818/j.cnki.1671-1637.2020.05.017 [21] LÜ L Y, ZHOU T. Link prediction in complex networks: A survey[J]. Physica A: Statistical Mechanics and its Applications, 2011, 390: 1150-1170. doi: 10.1016/j.physa.2010.11.027 [22] 户佐安, 邓锦程, 韩金丽, 等. 图神经网络在交通预测中的应用综述[J]. 交通运输工程学报, 2023, 23(5): 39-61. doi: 10.19818/j.cnki.1671-1637.2023.05.003HU Zuo-an, DENG Jin-cheng, HAN Jin-li, et al. Review on application of graph neural network in traffic prediction[J]. Journal of Traffic and Transportation Engineering, 2023, 23(5): 39-61. doi: 10.19818/j.cnki.1671-1637.2023.05.003 [23] 赵晓丽, 苏筠. 基于微博文本数据的灾害信息公众关注分析——以河南暴雨洪涝事件为例[J]. 自然灾害学报, 2024, 33(1): 89-98.ZHAO Xiao-li, SU Yun. Research on public concern of disaster information based on Weibo data: Take the rainstorm flood event in Henan Province as an example[J]. Journal of Natural Disasters, 2024, 33(1): 89-98. [24] YAN D Y, LI K P, GU S, et al. Network-based bag-of-words model for text classification[J]. IEEE Access, 2020, 8: 99. [25] YANG L, LI K P, ZHAO D, et al. A network method for identifying the root cause of high-speed rail faults based on text data[J]. Energies, 2019, 12(10): 1908. doi: 10.3390/en12101908 [26] 李曼, 刘贵源, 王艳辉, 等. 风险演化研究综述及其在轨道交通领域的应用与探讨[J]. 铁道运输与经济, 2025, 47(1): 13-30.LI Man, LIU Gui-yuan, WANG Yan-hui, et al. Review of risk evolution research and its application and discussion in rail transit[J]. Railway Transport and Economy, 2025, 47(1): 13-30. [27] PECH R, HAO D, PAN L M, et al. Link prediction via matrix completion[J]. Europhysics Letters, 2017, 117(3): 38002. doi: 10.1209/0295-5075/117/38002 [28] LI K P, GU S, YAN D Y. A link prediction method based on neural networks[J]. Applied Sciences, 2021, 11: 5186. doi: 10.3390/app11115186 [29] 潘黎明. 复杂网络上的链路预测及信息传播研究[D]. 成都: 电子科技大学, 2019.PAN Li-ming. Studies on link prediction and information spreading in complex networks[D]. Chengdu: School of Computer Science & Engineering, 2019. -
计量
- 文章访问数: 70
- HTML全文浏览量: 29
- PDF下载量: 17
- 被引次数: 0
