关键词: 盾构, 复合地层, 掘进参数, 预测, 卷积神经网络

Abstract: The effective prediction of tunneling parameters is the key to the intelligent improrement of shield. In order to establish the prediction model of shield tunneling parameters under soft and hard mixed ground, a quantitative method of geological conditions is proposed. Relying on the shield tunnel data of Jinan metro line R1, R2, and R3 under soft and hard mixed ground, the quantitative matrix of geological conditions is established by unitizing the stratum within the diameter of the tunnel and the upper and lower excavation faces along the buried depth, and counting the stratum characteristic parameters by units. Further, the characteristics of the geological quantitative matrix are automatically extracted based on convolutional neural network, and the prediction of tunneling parameters are finally realized. The results show that: （1）The prediction model of tunneling parameters based on geological quantification matrix performs well in training set, verification set, and test set. （2）When the model is applied to a new shield tunnel with similar geological conditions, the predicted values of cutter head torque, total propulsion force, propulsion speed and cutterhead speed are consistent with the actual values, and the average error is within 15%. （3）The prediction accuracy can meet the engineering requirements. Compared with the current BP neural network model based on weighted geological parameters, the prediction accuracy and stability of tunneling parameters have been significantly improved.

Key words: shield, mixed ground, tunneling parameters; prediction, convolution neural network