预测土饱和导水率的优化TabPFN-梯度提升多模型集成策略OA
Optimized TabPFN-gradient boosting multimodel ensemble strategy for predicting soil saturated hydraulic conductivity
土的饱和导水率是岩土工程和水文学中的关键水力参数,传统试验测定方法耗时费力,难以满足工程实践中快速应用的需求.本文基于易于测得的土壤基本性质构建预测模型,以颗粒组成质量分数和容重作为输入特征,对饱和导水率进行预测.提出了一种优化的表格先验数据拟合网络(TabPFN)模型,该模型融合了专家机制和两阶段优化策略,并将其性能与传统梯度提升模型和随机森林模型进行了比较.构建了多模型分层加权集成策略,以综合各类模型的优势.结果表明,专家机制和两阶段优化策略显著提升了Tab-PFN的性能.优化后的TabPFN模型最高预测精度R2为0.956 8,RMSE为0.001 1 cm/s;与优化前模型相比,R2提高了16.78%,且优于所有梯度提升模型和随机森林模型.分层加权集成策略能够有效整合各子模型在不同土壤质地条件下的预测优势,取得了最高的预测精度(R2为0.961 7,RMSE为0.001 0 cm/s),并有效降低了预测误差和异方差性.本文提出的优化集成模型为工程实践中快速估算饱和导水率提供了一种新的智能化方法.
Soil saturated hydraulic conductivity is a critical hydraulic parameter in geotechnical engineering and hydrol-ogy.Traditional experimental measurement methods are time-consuming and labor-intensive,making it difficult to meet the rapid application requirements of engineering prac-tice.This study constructs prediction models based on eas-ily measurable basic soil properties,using particle composi-tion percentages and bulk density as input features to predict saturated hydraulic conductivity.The study proposes an op-timized tabular prior-data fitted network(TabPFN)model that integrates an expert mechanism with a two-stage optimi-zation strategy and compares its performance with tradi-tional gradient boosting models and random forest(RF).A multimodel stratified weighted ensemble strategy is con-structed to synthesize the advantages of all models.The re-sults show that the expert mechanism and the two-stage opti-mization strategy significantly improve TabPFN perfor-mance.The optimized TabPFN model achieved a maximum prediction accuracy of R2=0.956 8 and root mean square er-ror(RMSE)of 0.001 1 cm/s,representing a 16.78%im-provement in R2 over the preoptimization model and outper-forming all gradient boosting and RF models.The stratified weighted ensemble strategy effectively integrated the predic-tion advantages of each sub-model across different soil tex-tures,achieving the highest prediction accuracy(R2=0.961 7,RMSE=0.001 0 cm/s)and effectively reducing prediction errors and heteroscedasticity.The proposed opti-mized ensemble model provides a new intelligent method for the rapid estimation of saturated hydraulic conductivity in engineering practice.
刘翼飞;倪钧钧;SADEGHI Hamed
东南大学交通学院,南京 211189东南大学交通学院,南京 211189Department of Civil Engineering,Sharif University of Technology,Tehran 1458889694,Iran
建筑与水利
土壤饱和导水率机器学习TabPFN梯度提升集成策略
soil saturated hydraulic conductivitymachine learningtabular prior-data fitted network(TabPFN)gra-dient boostingensemble strategy
《东南大学学报(英文版)》 2026 (3)
318-328,11
The National Natural Science Foundation of China(No.U2340227,52308342),Fundamental Research Funds for the Central Universities(No.RF1028623071,2242024k30066),Gradu-ate Research and Practice Innovation Program of Jiangsu Province(No.SJCX25_0128).
评论