Abstract: Accurate prediction of shallow underground temperature is of great significance for reducing investment risks and promoting the development and utilization of shallow geothermal energy. In this study, a hybrid model based on particle swarm optimization (PSO) and extreme gradient boosting (PSO-XGB) was developed and compared with single models including K-nearest neighbors (KNN), support vector regression (SVR), random forest (RF), and extreme gradient boosting(XGB). Firstly, 54 sets of borehole data were collected, and the dataset was expanded using Kriging interpolation. Latitude and longitude coordinates, annual average rainfall, annual average temperature, and distance to faults were used as input features for predicting the temperature at a depth of 100 meters underground. Then the performance of the models was evaluated using metrics such as root mean squared error, mean absolute error, coefficient of determination, andmean squared error. The results showed that the PSO-XGB hybrid model outperformed the single models in terms of predictive performance. The ERMS is 0.070 6, the EMA is 0.054 9, the R2 is 0.962 0, and the EMS is 0.005 0, which is significantly higher than the other models in terms of precision and degree of fitting. Therefore, the PSO-XGB hybrid model is superior to the single model in prediction performance.

Key words: underground temperature prediction, PSO-XGB hybrid model, KNN, SVR, RF, XGB