Identification of the formation temperature field by explainable artificial intelligence: A case study of Songyuan City, China

The development of the geothermal resource is crucial for advancing renewable energy and addressing global energy shortages. Despite the widespread application of artificial intelligence in geothermal research, limited focus has been placed on the interpretability of these models. Explainable artificial intelligence (XAI) models were applied in this study to predict geothermal temperatures, identify the primary influencing factors, and analyze the behavior of geochemical components. A novel unsupervised clustering-based group splitting method has been proposed for dataset segmentation. The SHapley Additive exPlanations (SHAP) method was then applied to quantify feature contributions and interactions, providing insights into the critical factors and mechanisms affecting geothermal temperature predictions. Leveraging model-derived insights, the geothermal field distribution in Songyuan City, China, was systematically analyzed. Finally, the study discussed the model's interpretative process and errors. The results revealed that the unsupervised clustering-based group splitting method significantly enhanced model performance, with depth and basement depth identified as the primary factors influencing geothermal temperatures. Areas with higher temperatures were concentrated near the basin center with shallower basement depths. This study first uses XAI to identify key factors and distribution patterns in geothermal temperatures, offering a valuable reference for applying XAI in geothermal research. • Explainable artificial intelligence (XAI) is applied to the geothermal temperature prediction. • An unsupervised clustering-based group splitting method for data partitioning is proposed. • XAI reveals the main influencing factors of geothermal temperature. • Knowledge embedding enhances the model's interpretability.