Prediction of Tunnelling Parameters for Underwater Shield Tunnels, Based on the GA-BPNN Method

Reasonable tunnelling parameters for underwater shield tunnels play an important role in maintaining driving efficiency and safety. In this paper, a neural network method was developed to predict tunnelling parameters. Soil properties and geometric parameters were taken as inputs for the neural network, which output the tunnelling parameters, such as advancing thrust, rotation, penetration, torque of the cutter head, and support pressure. In order to improve the stability of the neural network, a genetic algorithm (GA) with a global searching ability was used to optimize the initial weight of the neural network (GA-BPNN). The accuracy of the algorithm, based on GA-BPNN, was studied through an underwater shield tunnel project. The results showed that the integration of GA into the neural network significantly improves the prediction ability for shield tunnelling parameters, especially for adjustable parameters. Later, the developed GA-BPNN model was further utilized to predict and set the range of shield tunnelling parameters in fine sand stratum of high risk. Through a comparative analysis of tunnelling parameters, the reasons leading to ground instability have been found out, and the effectiveness of ground pre-reinforcement has been verified.