基于KPCA-TFPSO-BL的泥石流预测研究  

Study on debris flow prediction based on KPCA-TFPSO-BL

作　　者：徐根祺 曹宁 李璐 谢国坤[1] 党文博 Xu Genqi;Cao Ning;Li Lu;Xie Guokun;Dang Wenbo(School of Mechanical and Electrical Engineering,Xi'an Traffic Enginering Institute,Xi'an 710300,China;Civil Engineering College,Xi'an Traffic Enginering Institute,Xi'an 710300,China;School of Mechanical and Electrical Engineering,Tongchuan Vocational and Technical College,Tongchuan 727000,China;Municipal Road and Bridge Engineering Company,Shaanxi Construction Second Construction Group Co.,Ltd.,Baoji 721000,China)

机构地区：[1]西安交通工程学院机械与电气工程学院,西安710300 [2]西安交通工程学院土木工程学院,西安710300 [3]铜川职业技术学院机电工程学院,铜川727000 [4]陕西建工第二建设集团有限公司市政路桥工程公司,宝鸡721000

出　　处：《国外电子测量技术》2024年第10期81-90,共10页Foreign Electronic Measurement Technology

基　　金：陕西省自然科学基础研究计划(2023-JC-YB-464);西安交通工程学院中青年基金项目(2023KY-02)资助。

摘　　要：针对当前研究中泥石流诱发因子敏感度各异导致的预测准确度不高、数据集样本有限造成的模型训练和预测效果不佳、非线性过程严重引起的参数难以确定等问题,利用改进的核主成分分析算法(kernel principal component analysis,KPCA)筛选出相关性一般的因子,结合宽度学习(broad learning,BL)建立泥石流概率预测模型,再通过引入正弦因子的粒子群算法(TFPSO)对模型进行优化,最终建立基于KPCA-TFPSO-BL的泥石流预测模型。通过实验对比了经典BL模型、KPCAPSO-BL模型以及KPCA-TFPSO-BL模型的性能,结果表明,KPCA-TFPSO-BL的均方根误差为4.92,平均绝对误差为4.60,训练时间为7.22 s,该模型在预测误差和训练时间方面综合表现最佳。本研究为泥石流预测领域提供了一种新的思路和借鉴。In response to the problems of low prediction accuracy caused by the varying sensitivities of debris flow triggering factors in current research,poor model training and prediction performance due to limited dataset samples,and difficulty in determining parameters caused by severe nonlinear processes,an improved kernel principal component analysis(KPCA)algorithm was used to screen out factors with general correlation,combined with broad learning(BL)to establish a debris flow probability prediction model.Then,a particle swarm optimization(PSO)based on sine factors was introduced to optimize the model,and finally,a debris flow prediction model based on KPCA-TFPSO-BL was established.The performance of the classic BL model,KPCA-PSO-BL model,and KPCA-TFPSO-BL model was compared through experiments.The results showed that the root mean square error of KPCA-TFPSO-BL was 4.92,the average absolute error was 4.60,and the training time was 7.22 seconds.This model showed the best comprehensive performance in terms of prediction error and training time.This study provides a new approach and reference for the field of debris flow prediction.

关 键 词：泥石流 预测模型 宽度学习 

分 类 号：P642[天文地球—工程地质学] TN919.5[天文地球—地质矿产勘探]