遗传算法优化BP网络在滑坡灾害预测中的应用研究

在陕西省宝鸡市附近长寿沟地区滑坡详细调查和遥感解译的基础上，完成了1∶10000滑坡编目图。通过使用GIS的水文分析功能，运用正反DEM技术，将长寿沟地区划分为216个自然斜坡单元，其中包括123个滑坡单元和93个未发生滑坡单元，分析滑坡发生与坡高、坡度、坡向、坡形、人类工程活动和水文地质条件影响因子之间的统计规律。利用经遗传算法优化后的BP神经网络对80个滑坡样本和40个未滑坡样本进行训练学习，然后再利用训练好的网络对预测样本进行评价分析。结果表明:43个已滑坡单元中只有3个被误判为无滑坡，正确率为9302%，53个未滑坡单元中有10个被预测为滑坡，正确率为8113%，总体正确率为8646%。通过对被预测为滑坡的10个斜坡单元进行分析，发现这些单元在坡形、坡高等影响因素的组合上已经具备了发生滑坡的条件，虽然目前没有发生滑坡，但作为潜在的滑坡危险区，可以为滑坡灾害预测预报和防灾减灾工作提供参考。

Application of back propagation neural networks with optimization of genetic algorithms to landslide hazard prediction

Over the last decades,the development of Geographic Information System(GIS) technology has provided a method for the evaluation of landslide hazard.Through the use of direct-reverse DEM technology,the Changshougou valley is divided into 216 slope units,which includes 123 landslide units.According to mechanism analyses of landslides in the study area,six environmental factors are selected to evaluate the landslide occurrence,such as elevation,slope,aspect,curvature,distance to rivers,and human activities.Each factor is extracted in terms of slope unit within the scope of ArcGIS.The spatial analysis shows that most of landslides in the Changshougou valley are located at the elevation ranging from 100 to 150 m,with an aspect of 135°~225° and 40°~60° in slope,and on convex slopes,which are also influenced by hydrological processing and human activities.After the spatial analysis of environmental factors,this paper presents a case study for landslide hazard prediction,using back-propagation artificial neural network modeling optimized by genetic algorithms.Parameters of genetic algorithms and neural networks are set.The population size is 100,crossover probability 0.65,mutation probability 0.01,momentum factor 0.60,learning rate 0.7,max learning number 10000,and target error 0.000001.From a database of 216 landslides,120 landslides are used for training neural network models,and 96 landslides are used for the validation of landslide susceptibility.Comparing landslide presence with a susceptibility map,it is noted that the prediction accuracy of landslide occurrence is 93.02%,while the units without landslide occurrence is predicted with accuracy of 81.13%.The verification shows satisfactory agreement with accuracy of 86.46% between the susceptibility map and the landslide locations.In this case study,it is also found that some disadvantages can be overcome in the application of BP neural networks,for example,the low convergence rate and local minimum,after the optimization is carried out using genetic algorithm.In conclusion,we note that genetic algorithm-BP neural networks are an effective method to predict landslide hazard with high accuracy.