Hybrid support vector machine optimization model for inversion of tunnel transient electromagnetic method.

Area of Science:

• Geophysics
• Geotechnical Engineering
• Machine Learning

Background:

• Transient Electromagnetic Method (TEM) is vital for detecting adverse geological conditions in underground engineering.
• Accurate evaluation of geological features is critical for tunnel safety and construction.
• Existing 3D full-space TEM inversion for tunnels requires further development.

Purpose of the Study:

• To develop an advanced tunnel TEM inversion approach using a hybrid LSSVM and PSO model.
• To enhance the accuracy and reliability of TEM inversion for underground engineering applications.
• To address limitations in current 3D full-space TEM detection and inversion techniques.

Main Methods:

• Proposed a novel tunnel TEM inversion approach by integrating Least Squares Support Vector Machine (LSSVM) with Particle Swarm Optimization (PSO).
• Optimized LSSVM model parameters using PSO to mitigate randomness and uncertainty.
• Employed an orthogonal test method to refine initial PSO parameters for improved model accuracy.
• Conducted numerical simulations generating 125 datasets for model validation.

Main Results:

• The optimized PSO-LSSVM model demonstrated reduced randomness in PSO initial parameters and enhanced optimization.
• The model showed improved stability and accuracy in generalization ability compared to conventional methods.
• Numerical simulations confirmed the model's effectiveness in predicting TEM data values.
• Laboratory model tests validated the proposed method's efficacy.

Conclusions:

• The optimized PSO-LSSVM model offers an effective solution for tunnel TEM detection inversion.
• This hybrid approach significantly improves the accuracy and reliability of geological condition prediction.
• The study advances the application of machine learning in geophysical exploration for underground engineering.