基于INGO-RF的边坡稳定性预测模型  

Slope stability prediction model based on INGO-RF

作　　者：石峻峰 周琳 任宇联 王志鹏 SHI Junfeng;ZHOU Lin;REN Yulian;WANG Zhipeng(School of Civil Engineering and Architecture,Hubei University of Technology,Wuhan 430068,China)

机构地区：[1]湖北工业大学土木建筑与环境学院,武汉430068

出　　处：《安全与环境学报》2025年第4期1380-1390,共11页Journal of Safety and Environment

基　　金：桥梁结构健康与安全国家重点实验室开放项目(BHSKL18-03-KF)。

摘　　要：为提高边坡稳定性的预测精度以预防边坡失稳事故发生,提出了一种基于改进北方苍鹰算法优化随机森林(Improved Northern Goshawk Optimization algorithm optimized Random Forest, INGO-RF)的边坡稳定性预测模型。首先,根据413个边坡案例,选取重度γ、黏聚力c、内摩擦角φ、边坡角α、边坡高度H和孔隙压力比ru作为主要预测特征指标。其次,由于传统随机森林模型存在超参数问题,采用最佳值引导、减法优化器、柯西变异和动态调整搜索策略的INGO算法优化随机森林(Random Forest, RF)模型超参数。最后,与5种不同算法相比,所设计的INGO算法在8个测试函数中展现出更优的参数寻优能力和收敛速度;与5种不同预测模型相比,所设计的INGO-RF模型的各项评估指标均优于其他模型,该模型在训练集和测试集中的准确率分别为99.1%和91.2%,且发现γ是影响边坡稳定性的最敏感特征。研究表明,INGO-RF预测模型为边坡稳定性预测提供了一种新思路。This paper introduces a slope stability prediction model based on an Improved Northern Goshawk Optimization algorithm optimized Random Forest(INGO RF).Firstly,an original dataset was created from 413 slope cases,with 80%of the samples allocated for training and 20%for testing.Slope stability is primarily affected by factors such as soil shear strength,internal structure,and gravity.Key features selected for the model include bulk densityγ,cohesion c,internal friction angle∅,slope angleα,slope height H,and pore pressure ratio r u.Violin plots were generated to analyze the data distribution and assess the integrity of the dataset.Additionally,correlation matrix visualizations and calculations of Pearson correlation coefficients indicated weak or negligible correlations among the selected features.Secondly,to address the challenges of hyperparameter tuning in the random forest model,an Improved Northern Goshawk Optimization(INGO)algorithm was developed.This algorithm incorporates optimal value guidance and a subtractive optimizer in its initial stage to mitigate limitations such as blindness and the tendency to converge on local optima.In the second stage,a probability factor is introduced to facilitate Cauchy variation and dynamically adjust search boundaries for position updating strategies,thereby enhancing solution quality.Experimental results demonstrate that the proposed INGO algorithm outperforms five other algorithms in terms of parameter optimization capability and convergence speed across eight test functions.Notably,in multimodal functions f 13,the INGO algorithm achieves nearly an order of magnitude improvement in optimization accuracy compared to the classical Northern Goshawk Optimization(NGO)algorithm.Additionally,it reaches the theoretical optimal solution for single-modal functions f 1 after approximately 230 iterations.Compared to five other models,the INGO RF model excels in all evaluation metrics,achieving accuracies of 99.1%on the training set and 91.2%on the test set,demonstrating sensitivi

关 键 词：安全工程 边坡稳定性 可视化分析 改进北方苍鹰优化算法 随机森林 预测模型 

分 类 号：X43[环境科学与工程—灾害防治]