基于分形和神经网络相结合的土体冻胀量预测

根据土体的粒度分布具备分形性质的特征，通过理论分析和计算获得了所研究土体的分形维数，从而实现了土体结构特征的量化，为采用神经网络对冻胀量的预测过程中考虑士体的结构特征奠定了基础。在研究了BP神经网络的基础上，建立了其拓扑结构，采用L-M优化算法进行了迭代求解，预测结果与试验结果具有良好的一致性和吻合度，反映了土体冻胀过程的非线性特征和局部特征，弥补了理论模型和数值分析中无法考虑土体内部结构的缺陷，以及在预测中考虑土体的结构特征是必要的。

Prediction of frost heave in soil based on fractal and neural network

Based on granularity distribution of soil having fractal character, the fractal dimension of soil is studied by theory analysis and calculation. The structure character of soil is quantized by using fractal dimension, which lays a foundation for neural network considering soil structure in the process of prediction. Topology structure of BP neural network is built, and L-M arithmetic is used to find a solution. It has favorable coherence and curvature tolerance between prediction result and test result. This method remedies the defect of theory model and numerical analysis, which is unable to consider interior structure of soil. The research shows that it is essential to consider structure character of soil in the process of prediction,