基于数据驱动技术的收缩膨胀指数测定  

Measurement of Shrinkage and Expansion Index Based on Data-driven Technology

作　　者：王娟 WANG Juan(CRCC Harbour&Channel Engineering Bureau Group Co.,Ltd.,Wuhan 430000,China)

机构地区：[1]中国铁建港航局集团有限公司,武汉430000

出　　处：《价值工程》2025年第12期142-144,共3页Value Engineering

摘　　要：近年来,人工智能在岩土工程中的应用受到广泛关注,尤其是在预测膨胀土性能方面。然而,现有研究在膨胀土收缩-膨胀指数的预测精度方面仍存在不足。本研究提出了一种基于LightGBM模型的预测方法,并结合贝叶斯优化技术,以提高模型对膨胀土收缩-膨胀指数的预测精度。本文选取液限、塑限、塑性指数和线性收缩作为输入特征,通过贝叶斯优化对模型参数进行调优,显著提升了模型性能。实验结果表明,优化后的BO-LightGBM模型在训练集和测试集上的预测精度均优于未经优化的模型,其中测试集的预测准确度提高了约6%,模型拟合度(R^(2))提升了约5%,表现出较强的泛化能力和稳定性。该方法为岩土工程中膨胀土的性能预测提供了一种高效且可靠的解决方案,具有重要的实际应用价值。In recent years,the application of artificial intelligence in geotechnical engineering has received widespread attention,especially in predicting the performance of expansive soils.However,existing studies still lack accuracy in predicting the shrink-swell index of expansive soils.This study proposes a prediction method based on the LightGBM model and combines with Bayesian optimization to improve the prediction accuracy of the shrink-swell index.Liquid limit,plastic limit,plasticity index,and linear shrinkage were selected as input features,and Bayesian optimization was used to tune model parameters,significantly enhancing model performance.Experimental results show that the optimized BO-LightGBM model outperforms the non-optimized model in both training and test sets,with the prediction accuracy on the test set improved by approximately 6%and the model fit(R^(2))increased by about 5%,demonstrating strong generalization ability and stability.This method provides an efficient and reliable solution for predicting the performance of expansive soils in geotechnical engineering,with important practical value.

关 键 词：膨胀土 收缩-膨胀指数 机器学习 预测 

分 类 号：TU443[建筑科学—岩土工程]