摘要
针对传统静态机器学习模型在周期项位移预测中的缺陷和动态神经网络超参数人工选择困难的问题,在时序序列分解的基础上,提出一种新的滑坡预测耦合模型。首先,用最大相关最小冗余算法对周期项位移筛选合适的环境特征,作为长短期记忆人工神经网络的输入。然后,在天牛须搜索算法搜索过程中引入反馈机制,以避免原算法中出现远离最优解的问题;在算法迭代过程中将固定的递减因子改为动态递减因子,以提升前期全局和后期局部的寻优能力;利用改进的天牛须搜索算法对长短期记忆人工神经网络超参数进行寻优,以获得最佳的网络参数组合。最后,重构趋势项和周期项预测结果,得到最终预测位移。以发耳滑坡为例进行分析,结果表明:相较于其他方法,所提模型在平均绝对误差、均方根误差以及拟合度等方面更具优势。
In order to solve the problems of the defects of the traditional static machine learning model in the displacement prediction of periodic terms and the difficulty of manual selection of hyperparameters of dynamic neural networks,a new coupling model for landslide prediction was proposed on the basis of time series sequence decomposition.Firstly,the maximum correlation and minimum redundancy algorithm was used to screen suitable environmental features for the displacement of periodic items,which were used as the input of long short term memory artificial neural network.Then,a feedback mechanism was introduced into the search process of the beetle antennae search algorithm to avoid the problem of deviating from the optimal solution in the original algorithm.In the process of algorithm iteration,the fixed decreasing factor is changed to a dynamic decreasing factor to improve the optimization ability of the early global and late local optimization.The improved long term memory artificial neural network hyperparameters were optimized by using the improved beetle antennae search algorithm to obtain the best combination of network parameters.Finally,the prediction results of trend term and period term are reconstructed to obtain the final predicted displacement.Taking the Fa'er landslide as an example,the results show that compared with other methods,the proposed model has more advantages in terms of mean absolute error,root mean square error and fit.
作者
荆严飞
党建武
王阳萍
岳彪
JING Yanfei;DANG Jianwu;WANG Yangping;YUE Biao(School of Electronic and Information Engineering,Lanzhou Jiaotong University,Lanzhou 730070,China;Rail Transit Information and Control National Virtual Simulation ExperimentTeaching Center,Lanzhou Jiaotong University,Lanzhou 730070,China)
出处
《兰州交通大学学报》
CAS
2024年第2期58-67,共10页
Journal of Lanzhou Jiaotong University
基金
国家自然科学基金(61763025)
甘肃省教育科技创新项目(2021jyjbgs 05)
甘肃省军民融合专项(2020JG01)
甘肃省重点研发计划(21YF5GA158)
甘肃省知识产权计划项目(21ZSCQ013)。
