论文部分内容阅读
分析了前人关于融沉系数经验方法的研究结果,结果显示,与融沉系数关系最为密切的物性参数为液塑限、粉黏粒含量、干密度和含水量(含冰量).为了能够综合描述诸因素与融沉系数的经验关系,以兰州黄土和青藏黏土为试验对象,得到了两种具有不同物性参数的土在不同含水量和干密度条件下的融沉系数.采用BP神经网络算法对试验数据进行学习训练,得到了各因素与融沉系数间的经验关系数据库.为了提高训练样本的代表性,引用前人研究中的部分数据作为补充.对预留数据的预测结果表明,综合考虑多因素影响的BP神经网络经验方法具有较好的预测精度,而使用单一因素(含水量或干密度)预测融沉系数的经验方法其精度相对较差.
The results of previous empirical studies on thaw settlement coefficients were analyzed. The results show that the most close physical property parameters are the liquid-plastic limit, the powder clay content, the dry density, and the water content (including ice content) Based on the empirical relationship between the factors and thawing-thawing coefficient, two kinds of thawing and sinking coefficients of soils with different water content and dry density were obtained using Lanzhou loess and Qinghai-Tibetan clay as experimental objects. The BP neural network Algorithm is used to train the experimental data, and the empirical relation database between each factor and the thawing-sinking coefficient is obtained.In order to improve the representativeness of the training samples, some data in previous studies are supplemented.The prediction results of the reserved data show that, The BP neural network empirical method which considers the influence of multiple factors synthetically has better prediction accuracy. However, the empirical method using single factor (water content or dry density) to predict the thawing settlement coefficient is relatively poor in accuracy.