Optimising the use of energy and minimising carbon footprints in smart structures requires precise predictions of thermal energy usage. In this research, an innovative deep learning model that combines CNNs and RNNs is introduced as a data-driven strategy for short-term heat stress prediction. Incorporating made by hand elements like temporal properties and energy circulation variables, the framework identifies obscure trends from sensor information. homes, flats, and terraced homes are all part of the Danish housing structures dataset. Both the accumulated energy needs for different types of buildings and the usage for specific houses are taken into account in the two scenarios. Evaluation metrics including Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Coefficient of Determination (R3), and Mean Absolute Percentage Error (MAPE) are used to assess the suggested model on forecasts two days and one week in the future. According to the results, the CNN-RNN model is more accurate and resilient than the traditional models when it comes to estimating the demand for thermal energy. Smart energy handling is crucial for lowering usage and increasing cost efficacy, according to the study. This study improves smart structure sustainable by facilitating accurate forecasting, which in turn optimises heating equipment, allowing for more efficient allocation of power sources without sacrificing comfort for residents. The suggested CNN-RNN model has an overall accuracy of 95.0% in comparison to alternative models.