摘要
滑坡是我国西南地区一种典型的地质灾害,滑坡运动距离预测对于地震滑坡的致灾范围评估具有重要意义。对影响滑坡运动距离的因素进行了主成分分析,结合滑坡运动机理,将提取的3个主成分分别命名为动能因子、坡度因子和阻力因子,以此作为BP神经网络的输入神经元;利用遗传算法的全局搜索功能优化BP神经网络的初始权重和阈值,基于优化后的网络构建滑坡最大垂直运动距离和最大水平运动距离的预测模型;同时在主成分分析的基础上构建多元回归预测模型,并将网络模型与多元回归的预测结果进行对比,结果表明:遗传算法对于BP神经网络的优化效果明显,优化后的BP神经网络模型对于坡脚型滑坡运动距离的预测精确且稳定,最大水平和垂直运动距离预测误差在10%以内的分别占86.67%和93.33%,预测精度优于初始BP神经网络和多元回归预测模型。
Landslide is a typical geological disaster in southwest China.The prediction of the sliding distance is of great signi-ficance for estimating the disaster scope of landslide triggered by earthquake.The principal component analysis on the factors affecting the sliding distance was performed in combination with the sliding mechanism,where three principal components were selected and named as kinetic energy factor,slope factor and resistance factor,which were treated as the input to back propagation(BP)neural network.The initial weight and threshold of BP neural network were optimized by means of the global search function of genetic algorithm.The models for predicting both maximum vertical sliding distance and maximum horizontal sliding distance were constructed based on the optimized network.At the same time,the multiple regression prediction model was developed on the basis of principal component analysis,and the prediction results from network model were compared to those of the multiple regression.The results show that the effect of the genetic algorithm on the optimization of BP neural network was remarkable.Accurate and stable prediction of the sliding distance of the slope-foot landslide using the optimized BP neural network model can be achieved.After optimization the prediction deviation of both the maximum horizontal and maximum vertical sliding distance were less than 10%,accounting for 86.67% and 93.33% respectively,the accuracy of the prediction were higher than that based on BP neural network and multiple regression prediction model before optimization.
作者
胡俊涛
张细香
Hu Juntao;Zhang Xixiang(The Fifth Geological Survey Team of Hubei Geological Bureau,Huangshi Hubei 435004,China)
出处
《化工矿物与加工》
CAS
2022年第8期22-26,34,共6页
Industrial Minerals & Processing
关键词
坡脚型滑坡
运动距离预测
主成分分析
BP神经网络
遗传算法
预测模型
网络模型
slope-foot landslide
sliding distance prediction
principal component analysis
back propagation(BP)neural network
genetic algorithm
prediction model
network model
