论文部分内容阅读
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.
In order to find a way to indirectly obtain frost heave rate of seasonal frozen soil based on macroscopic and microscopic physical parameters, the parameters of soil quality can be easily tested according to the existing techniques and the frost heave rate of seasonal frozen soil is predicted by BP neural network. Four parameters of unit and structure unit and three parameters of external condition were used respectively to establish a neural network prediction model of frozen soil frost heave rate in season.The results showed that the maximum error was 0.19 and the minimum was 0.00 in 33 test samples , The error of four samples is between 0.1 and 0.19, the error of other samples is less than 0.05, accounting for 88% of the total samples, indicating that the model can reflect the basic trend of frost heave changes.Therefore, BP neural network frost heave rate prediction model of micro-physical parameters is feasible.