Prediction of Tunnel Squeezing in High Geostress Conditions Using SSA-LightGBM

Tunnel squeezing is a common geological hazard in high in-situ stress soft rock areas, and high-precision prediction of tunnel squeezing severity is of great significance. In view of the limitations of traditional binary classification models, a multi-class prediction model, SSA-LightGBM, optimized based on the Sparrow Search Algorithm (SSA), is proposed to achieve accurate multi-class prediction of the severity of tunnel compression. By collecting 143 tunnel engineering cases from both domestic and international sources, a database containing five key parameters for predicting large deformation is constructed. According to China's railway tunnel large deformation classification standards, the performance of the model is evaluated using a 10-fold cross-validation method. The results show that the average accuracy of the SSA-LightGBM model in predicting squeezing at all levels reaches 92.9%, which is 4.3% to 30.4% higher than that of the other seven mainstream machine learning models. The SSA-LightGBM model is applied to predict the large deformation sections of actual tunnel engineering, and the prediction results are basically consistent with the actual situation, verifying its practical engineering applicability.