基于软土常规物理参数的压缩模量预测研究

Prediction Study on Compression Modulus of Soft Soil Based on Conventional Physical Parameters

为建立软土常规物理参数与压缩模量之间的模糊数学关系并对土体压缩模量进行预测,以27个钻孔中90件高质量淤泥及淤泥质土试样的土常规物理参数及压缩模量数据作为研究的数据源,基于MLP和RBF神经网络分别以tanh、sigmoid和标准径向基函数、一般径向基函数为激发函数建立不同的“软土物理参数-压缩模量”神经网络预测模型。结果显示,通过优选可以得到对于Es1-2预测值和Es2-4预测值的MRE分别在5%左右和11%以下、RMSE分别在6%左右和14%以下预测模型。结论表明,在一定误差的允许范围内,基于神经网络可以依靠软土的常规物理参数对压缩模量进行预测。

In order to build the fuzzy mathematical relationship between the conventional physical parameters and the compression modulus of soft soil and predict the compressive modulus of soil,90 pieces of high quality mud and silt soil samples in 27 holes'physical parameters and compression modulus data were provided to the research.Different neural network forecasting models of "soft soil physical parameters and compression modulus"were built which were based on MLP and RBF neural network respectively to tanh,sigmoid and standard radial basis function and radial basis function as the excitation function.The results showed that Es1-2 predicted value and the predicted value of Es2-4 in MRE were about 5% and 11%,and the RMSE prediction model was about 6% and 14% respectively.Therefore,in the range of allowable error,the neural network could rely on the conventional physical parameters of soft soil to predict the compression modulus.