In order to study the correlation of key construction parameters for slurry balance shield and their correlation mechanism, based on the excavation data of 863 rings in shield section 3#~2# of Qinghuayuan Tunnel, normality of key parameters in different strata were judged through statistical methods and their relationships were systematically explored. Based on common machine learning algorithms such as random forest and BP neural network, the prediction model of cutterhead torque in complex interbedded stratum was established to reflect the priority of the model input characteristic quantity. The results show that the difference between the stratums has a significant effect on the correlation strength of excavation parameters, which can be changed from very weak correlation to strong correlation. The study indicates a strong correlation among cutterhead torque, slurry pressure and shield thrust in multiple strata. The input of weakly correlated parameters in the prediction model has a limited effect on the improvement of the prediction accuracy and the quality of the prediction model can be guaranteed by selecting some strongly correlated parameters.