BP神经网络在基坑周边地表短期沉降预测中的应用

Application of BP Neural Network to Short-term Predicting Foundation Settlement in the Base Pit Vicinity

在对基坑的监测数据进行预测和分析中,现有的一部分方法很难满足实际施工中高度非线性问题的拟合,如指数法预测的沉降量往往偏小,双曲线法预测的沉降量往往偏大,而GM(1,1)对观测值的累加往往又不具有指数规律。考虑到这些局限,引用BP神经网络,以苏州地铁2号线某工程为例,结合历史的沉降监测值,对其基坑周边地表短期沉降进行预测。实践表明,该方法预测误差较小,为基坑周边地表沉降的预测提供了一种较好的途径,在基坑动态设计与信息化施工方面具有重要的参考价值。

To make the engineering project safe and implement the information construction, it is important to ana- lyze and predict the monitoring data during pit excavation. Part of the current method is hard to meet the practical con- struction of highly nonlinear problem of fitting, such as by index method of settlement the prediction is often small, by hyperbolic method of settlement the prediction is often partial, and GM （ 1 , 1 ） on the observation value of the accumu- lator often does not have index law. Based on these considerations, the project which belongs to Suzhou metro line 2 projects is taken as an example. Using BP neural network knowledge, combined with the history of the settlement moni- toring value, the surface subsidence around the foundation pit is short-term predicted. Application results show that its prediction error is small. The propased method presented here is an important reference to the dynamic design and informative construction for pit engineering.