This paper introduces a method for predicting internal environmental parameters of the Yungang Grottoes using a hybrid approach that combines Long Short-Term Memory (LSTM) neural networks with Convolutional Neural Network (CNN) technology. To enhance the predictive accuracy, the Bayesian Optimization algorithm was employed to optimize the most relevant hyperparameters of the proposed model. The study utilized a database obtained from external weather stations and internal cave sensors as a sample set to train the model. Initially, a preprocessing algorithm was applied to improve data quality and prediction accuracy. Subsequently, the developed model was utilized to forecast the environmental parameters inside the cave under external conditions. The obtained results were compared with other model types: LSTM neural networks and CNN-LSTM hybrid neural networks. The findings indicate that the proposed model surpasses the comparative models in terms of prediction accuracy and quality across all analyzed parameters.