汕头海湾隧道复合地层超大直径泥水盾构掘进参数预测研究

为研究复合地层超大直径泥水盾构掘进参数之间的复杂关系，以汕头海湾隧道工程为背景，选取700环掘进数据，通过优选函数类别和网络结构，建立基于BP神经网络的复合地层超大直径泥水盾构掘进参数预测模型，定量预测刀盘转矩、刀盘能耗和平均泥水压力。研究表明： 1）复合地层盾构贯入度和掘进速度、贯入度和刀盘转速、刀盘电流和刀盘转速的皮氏积矩相关系数绝对值均在0.75以上，具有良好的线性相关性，其他掘进参数之间相关性较不明确； 2）复合地层盾构刀盘转矩和刀盘能耗预测值与实际值的算术平均误差在5%左右，平均泥水压力预测值与实际值的算术平均误差为1.31%，预测精度较高，满足盾构施工要求； 3）基于BP神经网络预测模型，软土地层各掘进参数预测效果得到进一步提升； 4）根据BP神经网络预测模型输入参数定量预测其他掘进参数，操作简单，预测效果良好，能够为盾构主司机提供参考，同时提高施工效率，为实现智能掘进打下基础。

Research on Prediction of Tunneling Parameters of Super large Diameter Slurry Shield in Composite Strata of Shantou Bay Tunnel

In order to study the complex relationships among tunneling parameters of super large diameter slurry shield in composite strata, a prediction model of tunneling parameters based on BP neural network and 700 ring tunneling data of Shantou Bay Tunnel Project is established. By optimizing function category and network structure, the prediction model can predict cutterhead torque, cutterhead energy consumption and average slurry pressure more accurately and quantitatively. The results show that: (1) In composite strata, the absolute value of the Pearson product moment correlation coefficient between penetration and tunneling speed, that between penetration and cutterhead rotation speed, and that between cuttehead current and cutterhead rotation speed are all above 0.75, which have good linear correlation; the relationships among other tunneling parameters are not clear. (2) In composite strata, the average errors of cutterhead torque and energy consumption between the predicted values and monitored data are about 5%, that of slurry pressure between predicted value and monitored data is 1.31, which indicates that the prediction results have high precision and can meet the requirements of shield tunneling. (3) The prediction results of tunneling parameters in soft soil stratum have been further improved by the prediction model based on BP neural network. (4) By quantitatively predicting other tunneling parameters in prediction model based on BP neural network, the operation is simplified, and the prediction effect is good, which can provide reference for shield driver, improve the construction efficiency and even realize intelligentized shield tunneling.