Prediction and Parameter Optimization of Surface Settlement Induced by Shield Tunneling Using Improved Informer Algorithm

Given the complex mechanism of surface settlement induced by shield tunneling, accurate prediction of surface settlement and rational design of tunneling parameters are critical for ensuring safe and efficient tunneling operations. To address the limitations of the Informer algorithm in predicting surface settlement during shield tunneling, the standard convolution was replaced with dilated causal convolution, and three measures were employed: soil layer classification and characterization, a moving prediction window, and special factor handling. An improved Informer algorithm was developed. The prediction time of the improved Informer algorithm was reduced by up to 30.64% compared to the original algorithm. The improved Informer algorithm exhibited significant enhancements in prediction accuracy, perception range across different timescales, and computational efficiency. Compared with random forest and long short-term memory algorithms, the improved Informer algorithm achieved superior performance, making it more suitable for predicting surface settlement induced by shield tunneling. By integrating the improved Informer algorithm with Shapley additive explanations theory, the contributions of shield tunneling parameters to surface settlement were analyzed. A multi-objective optimization algorithm was constructed for the shield tunneling parameters. Key parameters, including the shield thrust, cutterhead torque, and tunneling speed, were selected for optimization. The surface settlement after parameter optimization was reduced by 16.79%.