基于SSA-BP的泥石流敏感性分析

Debris Flow Susceptibility Analysis Based on SSA-BP

选择坡度、坡向、曲率、年降雨量、归一化植被指数、地层岩性、距构造距离、土地利用、居民密度以及路网密度这10个泥石流灾害的主控因素作为评价因子,采用麻雀搜索算法(Sparrow Search Algorithm,SSA)优化BP神经网络,并用SVM,SSA-SVM,BP,SSA-BP这4种机器学习模型评价云南省昆明市东川区的泥石流敏感性。结果表明:SSA-BP神经网络模型对东川区泥石流的预测成功率可以达到85.9%,相较于SVM,SSA-SVM,BP神经网络模型的预测准确率分别提高了3.9%,1.6%和3.3%。本文用泥石流灾害点对所生成的泥石流敏感性图进行验证,表明大部分的泥石流灾害点落在了敏感性为高和极高的区域内,证明了所生成的敏感性图对东川区的城乡规划、道路规划、防灾减灾等方面具有实际指导意义。

The main controlling factors of 10 debris flow disasters including slope, aspect, curvature, annual rainfall, normalized vegetation index, stratum lithology, distance from structure, land use, residents density and road network density are selected as evaluation factors.Using Sparrow Search Algorithm(SSA) to optimize BP neural network, and use SVM, SSA-SVM, BP, SSA-BP machine learning models to evaluate debris-flows susceptibility in Dongchuan District, Yunnan Province. The results show that the prediction success rate of the SSA-BP neural network model for the debris flow in Dongchuan area can reach 85.9%, which is 3.9%, 1.6%and 3.3% higher than that of the SVM, SSA-SVM and BP neural network models respectively, and verify the debris flow susceptibility maps generated by the four models with debris flow disaster points. The results show that most of the debris flow disaster points fall in areas with extremely high and high susceptibility. It proves that the generated susceptibility map of Dongchuan District has practical guiding significance for the urban and rural planning, road planning, disaster prevention and mitigation of Dongchuan District.