【摘要】 工程建设中，精确的土层剪切波速是一项重要的抗震分析参数。基于BP神经网络和随机森林等深度学习法可处理非线性问题的优势，以镇江地区某深度处的物理力学参数为输入，以剪切波速为输出，构建了深度学习的两种预测模型，对不同深度范围的土层的剪切波速进行预测。经过与同深度的实测剪切波速进行比较，表明两模型的剪切波速预测值与实测值间的最大误差均小于20%,其中随机森林模型最大误差低于15%,预测的剪切波速明显较经验估值法更接近实际，表明该预测模型可满足抗震分析要求，且在镇江地区具有一定的实用性。更多还原

【Abstract】 In the field of construction engineering, the precise shear wave speed values of soil layers are crucial parameters for seismic analysis. Making the most of the advantage that deep learning methods such as Back Propagation(BP) neural networks and Random Forest can handle nonlinear problems, two predictive models was developed in this study. The models take physical and mechanical parameters of a certain depth in the Zhenjiang area as input and use shear wave speed as the output, predicting the shear wave speed for soil layers at different depths. After a comparison with the measured shear wave speed at the same depth, the maximum errors between the predicted and actual shear wave speeds for both models are less than 20%. Notably, the Random Forest model has a maximum error less than 15%. This shows that the predicted shear wave speed is noticeably closer to reality compared to estimates from empirical formulas. Thus, the prediction models meet the requirements for seismic analysis and possess practical use in the Zhenjiang area.更多还原