基于季节冻土微观结构特征的神经网络冻胀率仿真预测

为了寻求基于宏观-微观物理参数间接得到季节冻土冻胀率的途径, 根据现有技术手段容易测试到土的性质参数, 利用BP神经网络法对季节冻土冻胀率进行预测. 选取微观孔隙参数及结构单元体参数各4个、 外部条件参数3个共11个参数, 建立季节冻土冻胀率神经网络预测模型. 结果表明: 在33个检验样本中, 误差最大为0.19, 最小为0.00, 有4个样本的误差在0.1~0.19之间, 其他样本误差都在0.05以下, 占总样本数的88%, 说明模型能反映冻胀变化的基本趋势. 因此, 文中建立的基于11个宏观微观物理参数的BP神经网络冻胀率预测模型是可行的.

Simulation and Prediction Model of Frost Heaving Ratio of Neural Network Based on Microstructure Characteristics of Seasonal Frozen Soil

In order to get a approach based on macro and micro physical parameters by which we can obtain frost heaving ratio of seasonal frozen soil indirectly,it is easy to get nature parameters of soil according to the existing technical ways,we can predict frost heave ratio using BP neural network method.A total of 11 parameters(4 parameters of micro-pore,4 parameters of micro-unit,and 3 parameters related with external conditions) are selected to build neural network prediction model for frost heaving ratio.The results show that the maximal error is 0.19,the minimum is 0.00 in 33 samples.4 samples have error between 0.1~0.2,other samples are all below 0.05,accounted 88% for total samples.It shows the model has been built can reflect the right trend of frost heave.Therefore,BP neural network predictive model based on 11 physical parameters is feasible for seasonal frozen soil.