<p>Shield tunnelling parameters serve as critical control indicators during tunnelling, directly influencing tunnelling efficiency and construction safety. However, in shield construction, the selection of tunnelling parameters usually depends on manual experience, which can easily lead to inefficiencies and high costs. In this study, an intelligent decision method based on light gradient boosting machine (LGBM) and non-dominated sorting genetic algorithm II (NSGA-II) is proposed to optimize the tunnelling parameters. LGBM-NSGA-II is a multi-objective optimization algorithm that takes penetration and cutterhead speed as key optimization variables, and uses tunnelling speed and energy as the objective function. Firstly, high-quality shield operation data is obtained through data preprocessing of the hard rock section in the Shantou Bay Subsea Tunnel. Subsequently, a regression prediction of the tunnelling speed is obtained by using the LGBM. On this basis, the LGBM model is used as the fitness function, and the NSGA-II is integrated to achieve multi-objective optimization of key tunnelling parameters. After the decision optimization, the average tunnelling specific energy is reduced by 5.68%, and the average tunnelling speed is increased by 29.92%. Furthermore, the application value of the optimization model proposed in practical engineering is verified by field data. By determining the feasible region and the optimization model, the recommended values of the cutterhead speed and penetration are obtained, and the control range of the decision is set. These results can provide scientific guidance for the future selection of tunnelling parameters.</p>