基于 Ant-Miner 的洪灾风险区划模型及应用

摘要

应用蚁群优化算法（Ant Colony Optimization，ACO）进行规则挖掘是一个新的研究热点。为解决指标变量与风险级别间非线性关系，提出一种基于蚁群规则挖掘算法（Ant-Miner）的洪灾风险区划模型。在 GIS 技术支持下，将该模型应用于北江流域洪灾风险区划实例中，结果表明：① Ant-Miner 模型可挖掘15条适合研究区的洪灾风险分类规则，这些规则以简单的条件语句形式表现，便于生成风险区划图；② Ant-Miner 模型测试精度（95.1％）高于相同条件下 BP 神经网络模型的精度（92.9％），表明其分类性能更好，对洪灾风险区划具有更好的适用性；③研究区高风险区主要集中于降雨量较大、地势平缓低洼、人口财产密集的地区，与历史洪灾风险情况较吻合，表明所构建的模型科学合理，可为流域洪灾风险评价提供了新思路。%Using Ant Colony Optimization (ACO)to mine rules is a research hotspot nowadays.This pa-per proposed a new zoning model of flood risk based on ant colony rule mining algorithm (Ant-Miner)to solve the non-linear relationship between index and flood risk grade.The model was used in the Beijiang River basin with the support of GIS technique.The assessment results show that ① 15 simple rules ex-pressed in the form of conditional statement were mined by the Ant-Miner model.The rules are appropri-ate for the study areas and can be easily used for generating a zoning map of flood disaster risk.② The test accuracy is 95.1% in the Ant-Miner model ,92.9% in BP neural network model,indicating that the discriminative capability and flood risk zoning applicability of the former is stronger than the latter.③The high risk areas identified by Ant-Miner are mainly located in the regions with large precipitation,flat and low-lying terrain and dense population and property.These areas match well with the submerged are-as of historical flood disasters,indicating that the Ant-Miner model is reasonable and practicable and can provide a new method for flood risk assessment.