Prediction method of surface settlement of rectangular pipe jacking tunnel based on improved PSO-BP neural network

Area of Science:

• Geotechnical Engineering
• Tunneling Engineering
• Computational Intelligence

Background:

• Rectangular pipe jacking tunnels require robust safety control, especially when crossing existing infrastructure like expressways.
• Predicting surface settlement is crucial for managing risks associated with tunneling operations.

Purpose of the Study:

• To develop and validate a novel prediction model for surface settlement in ultrashallow buried, large-section rectangular pipe jacking tunnels.
• To provide theoretical support for the safety control of such tunneling projects.

Main Methods:

• Established a Particle Swarm Optimization-Backpropagation (PSO-BP) neural network model incorporating an adaptive mutation method.
• Utilized an improved Particle Swarm Optimization (IPSO) algorithm with adaptive inertia weight and mutation particles to optimize hyperparameters.
• Validated the model using field monitoring data from an ultrashallow large cross-section rectangular pipe jacking tunnel project.

Main Results:

• The improved PSO-BP model demonstrated superior prediction stability compared to traditional BP, AWPSO-BP, and PWPSO-BP models, particularly during gentle surface subsidence with significant undulations.
• Predicted subsidence values closely matched actual field measurements.
• The model exhibited significantly improved accuracy and robustness in predicting surface settlement.

Conclusions:

• The proposed improved PSO-BP mixed prediction model offers a reliable and accurate method for assessing surface settlement in rectangular pipe jacking tunnels.
• This approach enhances safety control strategies for tunneling projects, especially those involving ultrashallow buried, large-section tunnels crossing critical infrastructure.