Abstract: Accurate prediction and coordinated optimization of shield tunneling parameters are crucial for ensuring construction safety and enhancing tunneling performance. To address the engineering challenge of simultaneously optimizing tunneling efficiency, energy consumption, and operational stability, this study, based on the Shenzhen-Dongguan Intercity Railway project, selects cutterhead penetration rate, tunneling specific energy, and total cutterhead power as key optimization objectives. Ten critical construction parameters are screened as inputs using the Relief algorithm, and a regression prediction model based on random forest optimized by a golden sine-cosine improved particle swarm optimization algorithm (GDPSO-RF) is developed. Furthermore, the GDPSO-RF model is utilized as a fitness function and integrated with three algorithms—non-dominated sorting genetic algorithm(NSGA)-Ⅱ, multi-objective evolutionary algorithm based on decomposition (MOEA/D), and reference vector guided evolutionary algorithm(RVEA)—to form a hybrid multi-objective optimization framework. This framework comprehensively explores the solution space and generates a Pareto front solution set covering multiple working conditions. Finally, the optimal construction parameter combination is selected by combining the coefficient of variation-grey evaluation method. The results demonstrate that: (1) The GDPSORF model performs excellently in predicting cutterhead penetration rate, tunneling specific energy, and total cutterhead power, withR2values of 0.927 8, 0.900 9, and 0.943 7, respectively, on the test set. (2) The three optimization algorithms—NSGA-Ⅱ, MOEA/D, and RVEA—exhibit significant regional complementarity in the objective space, providing diverse parameter choices for different construction requirements. (3) The obtained optimal parameter combination shows remarkable effectiveness in engineering applications. Compared to the pre-optimization average values, the optimal parameters increase the cutterhead penetration rate by 18.7%, reduce tunneling specific energy by 21.2%, and decrease total cutterhead power by 12.1%.

Key words: random forest, multiobjective optimization, cutterhead penetration, total cutterhead power, shield tunneling