土压平衡盾构掘进控制不当易引起地层变形过大或沉陷，导致盾构停机或损害邻近建（构）构物。除加固地层外如何通过调整掘进模式与优化掘进参数实现地层变形控制仍是尚未解决的盾构施工关键科学问题。本项目结合模型试验和离散元-有限差分耦合数值模拟研究土压平衡盾构掘进模式影响下掘进参数与地层位移的演变规律，揭示满仓与非满仓不同掘进模式下土压平衡盾构掘进力学传递机制，明确不同土舱渣土比例下的开挖面临界支护压力特征，探究不同掘进模式影响下盾构刀盘形式对土压平衡机理和地层位移的影响。结合工程实例大数据，采用深度学习神经网络算法挖掘不同掘进模式影响下盾构掘进参数与地层位移之间的多层网络非线性关联规则，基于遗传算法进行优化获取全局最优解，建立基于容许沉降控制理论的土压平衡盾构掘进参数智能预测方法。预期成果可为复杂环境下土压平衡盾构掘进参数与地层位移智能控制提供理论依据，对指导土压平衡盾构安全掘进具有广泛应用前景。

Improper control of shield operation will cause large ground deformation or collapse which lead to stoppage of shield machine or endanger the adjacent structures. Except reinforcing the ground, how to minimize the ground deformation through adjusting the shield boring mode and optimizing the boring parameters is a critical unsolved problem in the field of shield tunnelling. Based on scaled model test and coupled analysis of discrete element-finite difference method, this project will fist investigate the effect of shield boring mode on the performance of advancing parameters and ground movement. The tunneling mechanics and the characteristics of critical supporting pressure at the excavation face will be studied when the shield heading through different boring modes (full chamber or partial empty chamber). The effects of different shield cutter patterns on the earth pressure balance mechanism and ground movement will be examined as well. Then, by collecting comprehensive shield advancing parameters in the field, deep learning will be used to mining the multilayer nonlinear correlation between the shield advancing parameters and ground movements, considering the effects of EPB shield boring modes, the neural network will be optimized by a genetic algorithm to obtain a best solution. Furthermore, an intelligent method for predicting the EPB advancing parameters will be proposed based on the allowable settlement controlling theory. This work will provide theoretical basis for intelligent control of advancing parameters and ground movement and hence it has wide application prospect for the safety of tunnelling in complex ground with EPB.