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

Abstract

The transient electromagnetic method (TEM) can effectively predict adverse geological conditions, and is widely used in underground engineering fields such as coal mining and tunneling. Accurate evaluation of adverse geological features is a crucial problem that requires urgent solutions. TEM inversion is an essential tool in solving such problems. However, the three-dimensional full-space detection of tunnels and its inversion are not sufficiently developed. Therefore, combining a least-squares support vector machine (LSSVM) with particle swarm optimization (PSO), this paper proposes a tunnel TEM inversion approach. Firstly, the PSO algorithm is adopted to optimize the LSSVM model, thus overcoming the randomness and uncertainty of model parameter selection. An orthogonal test method is adopted to optimize the initial parameter combination of the PSO algorithm, which further improves the accuracy of our PSO-LSSVM model. Numerical simulations are conducted to generate 125 sets of original data. The optimized PSO-LSSVM model is then used to predict certain values of the original data. Finally, the optimization model is compared with conventional machine learning methods, and the results show that the randomness of the initial parameters of the PSO algorithm has been reduced and the optimization effect has been improved. The optimized PSO algorithm further improves the stability and accuracy of the generalization ability of the model. Through a comparison of different machine learning methods and laboratory model tests, it is verified that the optimized PSO-LSSVM model proposed in this paper is an effective technique for tunnel TEM detection inversion.