三次样条插值对人工神经网络预测软土固结精确度的影响

The Effect of Cubic Spline Interpolation on the Prediction of Soft Soil Settlement with Artificial Neural Network

软土固结会引起漫长的地基沉降。人工神经网络(ANN)是预测地基沉降的一种常用工具。为了进行预测,需要使用一定的前期沉降观测数据训练ANN。采用两类训练方法:一类是直接使用观测数据训练网络,这是普通方法;一类是在观测数据中借助三次样条插值(CSI)技术进行等时距插值,然后一并利用观测数据和插值训练网络。三次样条插值是通过求解三弯矩方程组,在曲线的非连续数据点之间形成填充数据的技术。借助Mat Lab的函数Spline,可以完成插值计算过程。结果发现,在不同固结阶段进行预测,引入CSI插值训练的网络预测准确度均高于直接用观测数据训练的网络。这一发现,对于工程实践具有重要意义。

The consolidation of soft soil may cause a long-term settlement. BP artificial neural network （BP-ANN） is a general technique for the prediction of settlement. Teaching the BP-ANN with a certain amount of settlement data is indispen- sible for the purpose of prediction. Two types of network teaching methods are employed ： one involving the use of pure meas- urement records, and the other involving the use of both the measurement records as well as data produced by cubic spline interpolation （CSI）. CSI is a technique for producing data between recorded data interval on a curve by means of solving tridi- agonal linear equation. The results show that the prediction by the latter is consistently better than the prediction with the former. The finding is of great practical significance in engineering.