【摘要】 本文采用人工神经网络 BP算法对深基坑开挖工程中的参数进行辨识。将某些现场实测值作为网络输入 ,土层物性参数作为网络的输出 ,通过有限元计算取得学习样本来训练网络 ,从而达到对深基坑开挖工程中的参数进行辨识的目的。同时 ,本文提出了将极大似然估计引入 BP学习算法中 ,可以考虑学习样本和网络输入 (现场实测值 )的误差 ,可以求得所辨识参数的可靠度。本文还对动态调整 BP学习算法的学习速率因子、冲量系数以加快网络学习速度的算法进行了研究 ,本文算例表明本文算法训练速率可比传统 BP算法快 10倍以上。更多还原

【Abstract】 A novel method based on ANN BP algorithm to perform parametric identification in deep foundation excavation is proposed in the paper. Taking in situ measurements (displacements, pore pressures, stresses etc.) as network input and parameters to be identified as network output, the network is trained with the samples obtained from FEM computation. With the introduction of maximum likelihood approach, the errors of both the samples and the network input (in situ measurements) can be considered in the identification procedure, and the reliability of the identified parameters can also be obtained. To make the BP learning more efficient, a family of algorithms that optimize the learning rate factor and momentum factor dynamically are also studied in the paper. The numerical results provided in the paper illustrate that the computational effort for the learning process can usually be reduced by more than 10 times as compared with the conventinoal BP algorithm.更多还原