Performance forecasts based on machine learning of geothermal heat exchangers

Geothermal energy occupies a significant role among renewable energy sources due to its constant and stable production capacity, unaffected by meteorological conditions. The ground-air heat exchanger (GAHE) systems can improve the comfort conditions of their environs by utilising stable energy sourced from the soil. This study utilises Extreme Gradient Boosting (XGB), Random Forest (RF), Decision Tree (DT), Support Vector Regression (SVR), and K-Nearest Neighbours (KNN) machine learning (ML) regression models to forecast the air outlet temperatures of GAHE systems. Before the performance forecasts of the ML regression models, 354 numerical data points were acquired from a CFD-based software program. In the second stage, the acquired raw data were standardized using standard scaler pre-processing. In the third stage, hyperparameter optimisation was conducted on the entire dataset with the grid search approach. The five-fold cross-validation method was employed using the optimal hyperparameters, and the models' performance was assessed independently at each stage. The performance efficiency of each ML model was evaluated using RMSE, MAE, MSE, and R2 metrics. The findings indicate that the XGB model outperformed others, achieving the highest accuracy and the lowest error rate, with an R2 of 0.99 and an RMSE of 0.61. The RF and DT models exhibited the highest success rates following the XGB model, with R2 values of 0.98 and 0.97, respectively. The SVR and KNN models had inferior predictive performance relative to the other models. Ultimately, SHapley Additive exPlanations (SHAP) analysis was conducted on the top-performing XGB model. This analysis visualises the contribution levels of the variables influencing the model's decision-making process, enhances the model's interpretability, and identifies the most significant variables in the prediction process.