| description abstract | The construction disturbance mechanism of rectangular pipe-jacking tunnels is more intricate than that of circular tunnels, leading to potential issues such as excessive accumulation and deformation of the surrounding formation, which can result in engineering disasters. However, there is currently a lack of reliable methods for predicting these disturbances. Machine-learning techniques have the capability to analyze the influence of multiple independent variables on a dependent variable, offering a new approach for predicting surface settlement in the construction of rectangular pipe-jacking tunnels. To address the sensitivity of existing machine-learning models to initial parameters, an improved particle swarm optimization (IPSO) method is employed. This method incorporates an adaptive mutation technique, adaptive inertia weight, and postoptimization method for mutant particles to enhance the particle size and determine the probability of obtaining the optimal value. By leveraging the strong mapping and nonlinear fitting abilities of the backpropagation (BP) algorithm, the IPSO-BP algorithm model is developed and compared with the BP, support vector machine, and random forest (RF) models using actual monitoring data. The findings indicate that in the presence of specific noise in the surface settlement data, the IPSO-BP prediction model demonstrates an enhanced accuracy of 26%, 25%, and 10% for the left amplitude. This approach can serve as a valuable reference for settlement prediction in similar projects. | |
