结合GIS与FNN-SGD的雷波县泥石流易发性评价

Evaluation of Debris Flow Susceptibility in Leibo County Based on GIS and FNN-SGD

针对四川省凉山彝族自治州雷波县的泥石流易发性进行评价,首先考虑雷波县的自然地理和气候特点等因素,通过ArcGIS对获取的数据进行处理,并按照雷波县行政区域绘制相应的评价因子分类图,使用频率比法分析评价因子对泥石流灾害的敏感度;然后采用SMOTE算法增加少数类样本,使正样本与负样本达到均衡;最后提出FNNSGD模型应用于雷波县的泥石流易发性评价,并根据模型结果绘制泥石流易发性分布图,采用ROC曲线作为评估指标。实验结果表明,与前馈神经网络、逻辑回归、随机森林模型相比,FNN-SGD模型的准确率更高、稳定性更强,更适用于雷波县泥石流易发性评价。

In Leibo county,Liangshan Yi autonomous prefecture,Sichuan province,the debris flow susceptibility was evaluated,first of all,considering factors such as the natural geography and climate characteristics of Leibo,by ArcGIS to obtain data and draw the corresponding evaluation factor in Leibo county administrative areas classification diagram,use frequency ratio method analysis of evaluation factors on the sensitivity of the debris flow disaster;Then,use SMOTE algorithm to increase a few samples,make positive sample and negative sample reach equilibrium.Finally,the FNN-SGD model was proposed and applied to the assessment of debris flow susceptibility in Leibo county.Ac⁃cording to the results of the model,the distribution map of debris flow susceptibility was drawn and the ROC curve was used as the evaluation index.Compared with the feedforward neural network,logistic regression and random forest model,the results showed that:FNN-SGD model has higher accuracy and stability,and is more suitable for the assessment of debris flow susceptibility in Leibo county.