摘要:
机器学习（ML）已被广泛应用于桩基工程建模，取得了巨大的成功。然而由于桩体受载发生位移至一定程度时，桩端阻力才会变得显著，故预测长桩的桩端阻力通常具有挑战性。为准确预测长桩的桩端阻力，筛选出力学效应、桩体特性和土壤特性三个关键方面的相关因素，提出一种结合多折交叉验证、混沌序列、星鸦搜索算法、稀疏贝叶斯算法和最大信息系数检验的混合模型框架，在提升预测准确性的同时增强模型的可解释性。选择越南胡志明市所采集的920个现实长桩和超长桩桩基工程数据作为基准数据集，以均方根误差、平均绝对误差和相关系数作为模型预测准确性的指标。结果比较表明，所提模型在点预测方面都优于现有ML模型的预测，多种指标的值都接近最优。同时，本文计算了桩端阻力多种影响因子的相关性强度结合现实工程经验极大丰富了关于模型内部计算的可解释性，对软土地基下长桩的设计研究具有深远意义。
Abstract:
Machine learning (ML) has emerged as a powerful tool in the modeling of pile foundation engineering, achieving considerable success. Nevertheless, the prediction of pile base resistance poses a challenge, particularly for long piles. This resistance becomes significant only once the pile has been sufficiently loaded and displaced. To enhance the accuracy of predicting pile base resistance in long piles, we identified three crucial factors: mechanical effects, pile properties, and soil properties. We then developed a novel hybrid model framework. This framework integrates multi-fold cross-validation, chaotic sequences, the nutcracker search algorithm, sparse Bayesian algorithms, and the maximum information coefficient test, thereby improving both the predictive accuracy and the interpretability of the model. We selected a comprehensive dataset comprising 920 realistic instances of long and super-long pile foundation engineering projects from Ho Chi Minh City, Vietnam, as our benchmark dataset. Model performance was evaluated using root mean square error, mean absolute error, and the correlation coefficient as metrics. The results indicate that the proposed model outperforms existing ML models in point prediction, achieving values close to the optimal for various indicators. Furthermore, this study quantifies the correlation strengths among multiple factors influencing pile base resistance, incorporating real engineering insights to significantly enhance the interpretability of the model's internal calculations. This advancement holds profound implications for the design and research of long pile foundations in soft soil conditions.
下载:
