BP神经网络在冻土路基变形预测中的应用

在青藏铁路冻土路基现场实测资料的基础上，用改进的BP神经网络建立起了路基变形与地温、路基高度和上限之间的非线性映射。对某一典型路基第30年的变形进行了预测，结果显示路基的累计融沉量在冻胀量的两倍以上。从绘制的路基变形过程曲线可以很清晰地看出路基一年中的变形趋势和冻胀融沉区间。在4月份以后，路基的变形由冻胀向融沉转变，变形与地温有很好的正比关系，但是当地温升高到一定值时，路基的融沉量便不再随着地温的升高而增大。路基的冻胀与地温的关系也有相似的规律，说明地温对路基变形的影响存在一个比较明显的区间，在这个区间范围内的温度对路基变形的影响较大，这为控制路基的病害提供了一个比较有价值的信息。

Application of BP neural network on deformation prediction for permafrost embankment

Based on the measured data of Qinghai-Tibet railway BP neural network for the mapping of ground embankment, a nonlinear system was built up using temperature, embankment height and permafrost table with embankment deformation. The deformation in the 30th year of the embankment was predicted by using the trained BP neural network. Result shows that the accumulative thawing settlement is twice more than the frost heave. The embankment soil is with the rise of the changing ground from freezing status to thawing after April. The thawing settlement increases gradually temperature at the beginning and keeps steady after the temperature exceeding a certain value. The frost heave of the embankment has similar rule as the thawing settlement. It indicates that the ground temperature influences on the embankment deformation dramatically in a special range, and beyond this range, the influence is negligible. This can be valuable information for controlling the embankment destruction.