基于PSO-SVM的鹤盛溪流域山洪风险评价
Risk assessment of torrential floods in the Heshengxi Watershed based on PSO-SVM
王如锴 1练继建 2苑希民 1田福昌 1陈隆吉 3马文豪1
作者信息
- 1. 天津大学水利工程仿真与安全国家重点实验室,天津 300350;天津大学建筑工程学院,天津 300350
- 2. 天津大学水利工程仿真与安全国家重点实验室,天津 300350;天津大学建筑工程学院,天津 300350;天津理工大学海洋能源与智能建设研究院,天津 300384
- 3. 温州市水文管理中心,浙江 温州 325000
- 折叠
摘要
为探究温州市鹤盛溪流域山洪风险空间分布,综合考虑山洪致灾因子、孕灾环境和承灾体3 方面的山洪影响因子,建立基于粒子群优化-支持向量机(PSO-SVM)混合算法的山洪风险评价模型.选取准确度、灵敏度、特异性、F-score值、Kappa系数和受试者工作特征曲线等 6 个指标,采用学习矢量量化(LVQ)算法量化山洪影响因子对山洪灾害发生的影响程度,并将PSO-SVM混合算法模型与单一算法模型进行对比.结果表明:混合算法具有一定的迁移能力,能够更加准确地反映山洪风险的空间分布特征;验证集受试者工作特征曲线指标、Kappa 系数和准确度分别为0.934、0.833、0.912,PSO-SVM混合算法模型能显著提高山洪风险评价精度.
Abstract
To explore the spatial distribution of torrential flood risk in the Heshengxi Watershed of Wenzhou City,a risk assessment model of torrential flood based on particle swarm optimization-support vector machine(PSO-SVM)hybrid algorithm was established,taking into account the torrential flood causing factors,the disaster-prone environment,and the hazard bearing bodies.Six evaluation indicators were selected,including accuracy,sensitivity,specificity,F-score value,Kappa coefficient,and subject working characteristic curve.The learning vector quantization algorithm was used to quantify the impact of torrential flood impact factors on the occurrence of torrential flood disasters,and the PSO-SVM hybrid algorithm model was compared with single algorithm models.The results show that the hybrid algorithm has a certain transfer ability and can more accurately reflect the spatial distribution characteristics of torrential flood risk.The working characteristic curve indicators,Kappa coefficient,and accuracy of the validation set subjects were 0.934,0.833,and 0.912,respectively.The PSO-SVM hybrid algorithm model can significantly improve the accuracy of torrential flood risk assessment.
关键词
山洪灾害/风险评价模型/山洪影响因子/PSO-SVM/鹤盛溪流域Key words
torrential flood disaster/risk assessment model/torrential flood impact factors/PSO-SVM/Heshengxi Watershed引用本文复制引用
基金项目
国家重点研发计划项目(2022YFC3202501)
水利部重大科技项目(SKS-2022002)
科技部重点领域创新团队项目(2014RA4031)
国家自然科学基金委创新团队项目(51621092)
出版年
2024
国家科技期刊平台
NSTL
万方数据