软土盾构施工地表变形的小样本进化神经网络预测

A study of prediction of ground deformation with small samples in shield construction in soft clays using ENN method

上海地区软土具有高压缩性和易塑流等特性,在盾构机的挤压和不当施工扰动下将会引起较大的土层移动和地表隆陷。以盾构施工实测位移资料为学习样本,通过遗传算法搜索具有最优预测效果的神经网络结构及学习参数。利用获得的进化神经网络在小样本训练情况下建立模型,对下一步施工的地表变形进行合理的预测。对上海市某盾构隧道的施工地表变形预测表明该模型可获得较高预测精度。

Concerning the property of high compression and easy rheology of soft clay in Shanghai, a great movement and hogging or hagging of ground may be produced during shield construction due to extrusion and unreasonable disturbance. A new method is proposed to predict the deformation by evolutionary neural network(ENN). The Genetic Algorithms is used to study the optimal structure and parameters of ENN. Learning from small samples, the deformation of next step is obtained by ENN. The case study shows the method makes a good performance in the prediction of ground deformation.