基于集成因子优选算法的面板堆石坝变形预测模型  

Deformation prediction model of panel rockfill dam based on integrated factor optimization algorithm

作　　者：郑好[1] 李登华[2,3] 丁勇 ZHENG Hao;LI Denghua;DING Yong(School of Physics,Nanjing University of Science and Technology,Nanjing,Jiangsu 210094,China;Nanjing Hydraulic Research Institute,Nanjing,Jiangsu 210029,China;Key Laboratory of Reservoir and Dam Safety,Ministry of Water Resources,Nanjing,Jiangsu 210029,China)

机构地区：[1]南京理工大学物理学院,江苏南京210094 [2]南京水利科学研究院,江苏南京210029 [3]水利部水库大坝安全重点实验室,江苏南京210029

出　　处：《排灌机械工程学报》2025年第2期178-186,共9页Journal of Drainage and Irrigation Machinery Engineering

基　　金：国家重点研发计划项目(2022YFC3005502);国家自然科学基金资助项目(51979174);国家自然科学基金联合基金资助项目(U2040221);央级公益性科研院所基本科研业务费专项资金资助项目(Y321004)。

摘　　要：面板堆石坝变形主要受外部荷载和内部材料蠕变影响,而影响因子过多会造成特征冗余,引起预测精度过拟合问题;因子过少导致信息表达不全,造成预测效果差,模型泛化能力不足等问题.因此构建涵盖面广,可解释性强的因子集并从中优选格外重要.针对上述问题,文中提出基于集成因子优选算法的面板堆石坝变形预测模型,使用ReliefF与沙普利加性解释(SHAP)算法通过权重集成获得因子贡献值排序,再分析因子累计贡献率差量阈值剔除非关键因子,获得关键因子.以新疆某混凝土面板堆石坝为例,以特征缩减率FRR、归一化平均绝对百分比误差nMAPE、平均绝对误差MAE、均方误差MSE以及决定系数R^(2)为评价指标,试验结果表明文中提出的算法面对不同的预测模型都可准确地获得最佳因子,有效地提升了预测精度.相对于传统因子优选算法适应性更强,预测能力提升最显著,解决了影响因子冗余或者欠缺带来的预测能力较低的问题,提高了模型的泛化能力,为大坝安全监测研究提供了行之有效的因子优选方法.The deformation of panel rockfill dams is influenced by external loads and internal material creep.The presence of an excessive number of influencing factors can lead to feature redundancy and result in overfitting,thereby compromising prediction accuracy.Conversely,a lack of sufficient factors may lead to incomplete information representation,causing poor predictive performance and limited model generalization ability.Hence,it is crucial to build a comprehensive and interpretable set of factors to optimize them accordingly.To address these challenges,a predictive model for the deformation of panel rockfill dams utilizing an integrated factor optimization algorithm was introduced.ReliefF and Shapley additive explanations(SHAP)algorithms were employed to rank the importance of factors through weighted integration.Subsequently,non-essential factors were eliminated based on an analysis of the threshold of cumulative contribution rate differences,leading to the identification of key factors.By taking a concrete panel rockfill dam in Xinjiang as the research object,the feature shrinkage rate(FRR),normalized mean absolute percentage error(nMAPE),mean absolute error(MAE),mean square error(MSE),and coefficient of determination R^(2)were used as the evaluation indexes.The experimental results show that the algorithms proposed in the paper can accurately obtain the best factors in the face of different prediction models,which can effectively improve the prediction accuracy.Compared to conventional factor optimization techniques,the proposed approach exhibits greater adaptability and delivers more significant predictive enhancements,effectively addressing issues related to inadequate prediction capability caused by either redundant or insufficient influential factors.Furthermore,it enhances the generalization capacity of the model and offers an efficient method for optimizing factors in research pertaining to dam safety monitoring.

关 键 词：大坝安全监测 因子优选 SHAP理论 贡献度排序 

分 类 号：S277.9[农业科学—农业水土工程]