摘要
为提高降雨型滑坡的预测准确率,克服现有预测方法难以处理多因素非线性关系的问题,提出了一种基于DBN的对学习率进行优化控制的多因素预测模型,该模型在传统的DBN算法基础上,引入动量学习率,使用Dropout和Softmax等技术,避免收敛困难或局部最优,减少过拟合问题。仿真实验结果验证了本文所提出模型的准确率,为利用深度学习方法进行降雨型滑坡预测提供了新思路。
In order to improve the prediction accuracy of rainfall-type landslides and overcome the problem that the existing prediction methods are difficult to deal with multi-factor nonlinear relationships,a multi-factor prediction model based on DBN for optimal control of the learning rate is proposed.On the basis of the algorithm,the momentum learning rate is introduced,and techniques such as Dropout and Softmax are used to avoid convergence difficulties or local optimality,reduce overfitting problems.The simulation experiment results verify the accuracy of the model proposed in this paper,providing a new idea for using depth learning method to predict rainfall induced landslides.
作者
夏旭
谭韶生
Xia Xu;Tan Shaosheng(Hunan Vocational College of Safety Technology,Changsha,China;Hunan Industrial Vocational and Technical College,Changsha,China)
出处
《科学技术创新》
2023年第9期67-71,共5页
Scientific and Technological Innovation
基金
湖南省应急管理厅2020年度科技项目(2020YJ008)。
关键词
深度学习
滑坡
DBN算法
动量学习率
预测
deep learning
landslide
DBN algorithm
momentum learning rate
prediction