摘要
在甘肃黑方台,由于台塬大面积提水灌溉导致台塬边滑坡频繁发生,对当地居民生命财产安全造成危害。文章以黑方台75处滑坡(含53处黄土滑坡与22处黄土-基岩滑坡)为研究对象,基于BP神经网络模型,优化建立了迭代BP神经网络模型,全方位研究分析了滑坡滑距同滑坡体体积参数与地形地貌参数之间的关系,其中黄土滑坡迭代99次,黄土-基岩滑坡迭代256次。最优模型返回相关系数为0.976(黄土滑坡)与0.978(黄土-基岩滑坡),结果具有较高的可靠性。选取模型结果最优的几组迭代方案分析可知,影响黄土滑坡滑距的重要参数为滑坡体积、滑源区纵向长、滑坡落差、滑坡坡角、黄土厚度,次要参数为滑源区横向宽与滑坡壁高;影响黄土-基岩滑坡滑距的重要参数为滑源区纵向长、滑坡落差、滑动方向、基岩倾角、滑坡坡角,次重要参数为滑坡体积与岩层倾角,次要参数为滑源区横向宽与滑坡壁高。该研究为黑方台地区滑坡滑距以及滑坡危险范围的研究提供参考与指导。
Landslides developed frequently on the edges of the Heifangtai loess platform because of over irrigation in this area,which is considered to be a severe threat to the local community. In this paper,we took53 loess landslides and 22 deep-seated landslides as the research objects,and established an iterative program based on the BP neural network. Using the program,we modeled a comprehensive system of analyses and evaluating enterprises to analyze the internal relation between travel distance and slide volume landform of the slide point. Results of high reliability were obtained with the model. The relation coefficient is 0. 976 for loess landslide via 99 iterations and is 0. 978 for deep-seated landslide via 256 iterations. The results show that for loess landside,the most important features of travel distance are the slide volume,slide length,vertical drop,loess thickness and slope angle,the second important features are the slide width and slide thickness. For deep-seated landslide,the most important features of the travel distance are the slide length,vertical drop,travel direction,rock dip angle and slope angle. The second important features are the slide volume and rock dip direction angle,and the third important features are the slide width and slide thickness. This measure can give clear interpretation of the study of travel distance and the area of having disasters in Heifangtai region.
出处
《水文地质工程地质》
CAS
CSCD
北大核心
2016年第4期141-146,152,共7页
Hydrogeology & Engineering Geology
基金
国家重点基础研究计划(973计划)资助项目(2014CB744703)
国家杰出青年科学基金(41225011)
教育部"长江学者奖励计划"(T2011186)
关键词
滑坡灾害
滑距
迭代BP神经网络
评价参数
landsides
travel distance
iterative program based on the BP neural network
evaluation parameters