Abstract: To realize accurate prediction and real-time optimization control of large-diameter slurry shield tunneling performance under four complex geological conditions, including clay, fine sand dry, gravelly soil and strongly weathered mudstone, a real-time optimization control method for tunneling performance combining long short-term memory (LSTM), Kolmogorov-Arnold network (KAN), and multi-objective evolutionary algorithm based on decomposition (MOEAD) is proposed, which is based on a certain line of Wuhan Yangtze River Basin railway transportation. Initially, the overturning moment, over-under excavation volume and advance speed are taken as the optimization objectives, and the key construction parameters are identified and are taken as the input parameters by importance analysis and correlation analysis. Subsequently, the nonlinear mapping relationship between the optimization objectives and the key construction parameters is captured via LSTM-KAN to derive the fitness function of the optimization objectives. The interpretability of the model is improved via SHapley Additive exPlanations (SHAP), and key construction parameters are identified under four complex geological conditions. Ultimately, tunneling performance is optimized and controlled in real time via LSTM-KAN-MOEAD and technique for order preference by similarity to an ideal solution (TOPSIS). The results show that: 1) LSTM-KAN realizes the accurate prediction of tunneling performance under four complex geological conditions, and the MAPE is only 2.50% at the highest. 2) The key construction parameters are identified via SHAP as the total thrust, cutter torque, cutter rotational speed, and cutter squeezing force, respectively. 3) LSTM-KAN-MOEAD realizes the effective optimization and control of tunneling performance under four complex geological conditions, and the average overall improvement is as high as 54.66%. (4) Based on the optimization control results, the control range of key construction parameters under four geological conditions are proposed. The real-time optimization control for large-diameter slurry shield tunneling performance is achieved , and valuable guidance is proposed in this paper.