基于GA-BP神经网络的边坡变形预测  

Slope deformation prediction based on GA-BP neural networks

作　　者：谭文辉 李凯 刘慧敏 蔡美峰[1] 郭奇峰[1] TAN Wenhui;LI Kai;LIU Huimin;CAI Meifeng;GUO Qifeng(School of Civil and Resources Engineering,University of Science and Technology Beijing,Beijing 100083,China)

机构地区：[1]北京科技大学土木与资源工程学院,北京100083

出　　处：《工程科学学报》2025年第4期594-605,共12页Chinese Journal of Engineering

基　　金：国家重点研发计划课题资助项目(2023YFC2907302);国家自然科学基金资助项目(52274072)。

摘　　要：露天矿山高边坡的变形预测是保障矿山安全生产的重要手段.本文以西藏某矿山边坡为对象,采用高精度合成孔径干涉雷达对矿区南帮边坡进行了全天候位移监测,分析了边坡变形的基本规律;采用小波降噪理论对采集的时序位移监测数据进行了降噪处理,并且为了避免预测模型陷入局部极小值,引入遗传算法(即GA算法)整合进BP神经网络的训练步骤中,用于优化BP神经网络的初始权值和阈值设置,建立了GA-BP神经网络边坡变形时序预测模型,并与BP神经网络边坡变形时序预测模型进行对比分析.研究结果表明:GA-BP模型较BP模型的预测精度提高了10%以上,预测的平均误差减少了50%以上,而且预测的边坡变形趋势与监测值吻合程度更高;GA-BP模型较BP模型收敛速度加快10倍以上,GA-BP模型的回归系数、模型适应度优于BP模型.因此,采用GA-BP模型可使边坡变形预测的精度、收敛速度、泛化能力均得到提高,预测结果更为可靠,可为矿山边坡安全生产提供保障.Predicting deformation on high slopes is crucial for ensuring the safety of open-pit mining operations.Traditionally,empirical methods and numerical simulations have been employed to predict slope displacement.However,with advancements in artificial intelligence,machine learning has emerged as an important method for predicting slope deformation in open-pit mines.Currently,a popular approach is using the backpropagation(BP)neural network to construct a time-series deformation prediction model for slopes.To enhance the BP neural network performance and prevent it from falling into local minima,a genetic algorithm(GA)is introduced to the training step of the BP neural network.This algorithm optimizes the initial weights and thresholds of the BP neural network,leading to the establishment of the time-series deformation prediction model of slopes based on the GA-BP neural network.In this paper,we collected time-series displacement data of slopes from an open-pit mine in Tibet using high-precision synthetic aperture radar(SAR)for all-weather displacement monitoring.The wavelet denoising theory was also used to eliminate interference from abnormal displacement data.The structural parameters of the BP neural network were determined using the grid search method after comparisons.For model validation,we developed two slope time-series deformation prediction models:one using the GA-BP neural network and the other using the BP neural network.We evaluated their prediction effectiveness by examining the RMSE of predicted values,the number of training operations,and model adaptability to training and validation sets.In addition,we forecasted slope deformations at five monitoring points over the next 10 h using test set data not included in the model training.The results show that the GA-BP model achieves over 10%higher accuracy than the BP model in training and validation sets.It also converges>10 times faster and adapts better to model conditions.The maximum average error of the GA-BP model on the test set is only 2.48%,with

关 键 词：露天矿边坡 变形预测 BP神经网络 遗传算法 时间序列 

分 类 号：TD176[矿业工程—矿山地质测量]