基于机器学习方法的上海市暴雨内涝灾情预测模型研究

A study on rainstorm waterlogging disaster prediction models in Shanghai based on machine learning

本文基于区域灾害系统理论,综合考虑致灾因子、暴露度和脆弱性,提出了一套暴雨内涝灾情预测指标体系;在此基础上利用上海市应急联动中心110接报暴雨内涝灾情数据,构建了暴雨内涝灾情预测BP模型(Back Propagation Model)和XGBoost模型(Extreme Gradient Boosting Model),并对比分析了预测模型效果,实现对上海市暴雨过程内涝灾情数量预测;最后对内涝灾情影响等级进行阈值划分,以期为暴雨内涝影响预报与风险预警业务、服务及灾害管理提供技术支撑。结果表明:1)综合考虑致灾因子、暴露度、脆弱性指标且不经主成分分析降维的指标组合作为暴雨内涝灾情预测指标体系时,BP模型和XGBoost模型的预测精度最优;2)全量样本XGBoost模型总体表现最优,暴雨内涝灾情的右偏分布和内涝灾情的异常高值均对XGBoost模型预测误差有不同程度的贡献;3)综合评价法在历史灾情百分位法和模拟灾情百分位法基础上,结合多年业务实践经验和用户对于110灾情的处置承受力对暴雨内涝灾情进行阈值划分,在实际应用中具有一定参考意义。

Based on the regional disaster system theory,this paper comprehensively considers the hazard,exposure and vulnerability,and puts forward a set of rainstorm waterlogging disaster forecasting index system.On this basis,it constructs the BP model(Back Propagation Model)and XGBoost model(Extreme Gradient Boosting Model)for the rainstorm waterlogging disaster prediction,comparatively analyzes the prediction models effect by using the rainstorm waterlogging disaster reports which originate from Shanghai Emergency Linkage Platform 110 alert phone calls,and realizes the quantitative prediction of the rainstorm waterlogging disaster.Finally,it classifies the impact-level of the rainstorm waterlogging with a view to provide technical support for rainstorm waterlogging impact-based forecast and risk-based warning services and disaster management.The results indicate that:1)Comprehensively considering the hazard,exposure and vulnerability as the rainstorm waterlogging disaster prediction index system without principal component analysis,the prediction accuracy of BP model and XGBoost model is optimal;2)The overall performance of the XGBoost model based on the full sample is the best,the right-skewed distribution and the abnormally high value of the rainstorm waterlogging disaster numbers have important contribution to the prediction error of the XGBoost model;3)The comprehensive evaluation method divides the rainstorm waterlogging disaster threshold based on the historical disaster percentile and the simulated disaster percentile,and combines with professional practical experience for the 110 alert phone calls from the Shanghai Emergency Linkage Platform,it has certain reference significance in practical application.