Fluctuations in temperature and relative humidity within ancient tombs can lead to significant degradation of mural artworks, including salt efflorescence, weathering, delamination, and other forms of deterioration, thereby severely compromising their cultural and historical value. In the case of the Southern Tang Dynasty (Shunling Tomb) in Nanjing, Jiangsu Province, China, which are open to public visitation, rapid and accurate prediction of the indoor temperature and humidity is essential for the preventive conservation of these murals. However, conventional models based on heat-moisture coupling physics often suffer from inefficiency and lack of real-time applicability. To address these limitations, this study proposes an interpretable, high-performance machine learning-based model for real-time prediction of indoor environmental conditions in tombs. This model aims to meet the preventive conservation requirements of significant cultural heritage sites, as well as optimize visitor access management and environmental control. In this study, several machine learning algorithms are employed to predict temperature and relative humidity within the Southern Tang Tombs, using hourly outdoor meteorological data as input. Model performance is evaluated based on accuracy and precision, with SHapley Additive Explanations (SHAP) used to interpret the model outputs and quantify the influence of different input parameters. The findings indicate that the XGBoost model exhibits the highest accuracy in predicting indoor temperature, while CatBoost is more effective for real-time humidity prediction. SHAP analysis identifies soil cover properties and ambient thermal parameters as the governing factors in tomb microclimate regulation. The developed machine learning framework provides an efficient and accurate analytical solution for cultural heritage microclimate studies. Conservation strategies should focus on optimizing overburden configuration and implementing real-time thermal environment monitoring systems.