This research describes a novel technique for anticipating unstable heat transfer in porous media. Convolutional neural networks (CNNs) are used with finite volume method (FVM) and long short-term memory (LSTM) networks to accomplish this. Heat transport networks are difficult to characterise using traditional numerical methodologies owing to their nonlinearity and complexity. The proposed solution combines FVM's precise physical modelling with CNN's and LSTM's superior pattern identification and temporal analysis. This collaboration supports the suggested strategy. Heat transport dynamics simulations in porous materials are more accurate, efficient, and adaptable when employing this hybrid framework. The experimental setup focused on porous material properties and gathered and processed a large amount of data. The building's three-dimensional shape, heat transfer, and time were investigated. Temporal fluctuations were also used. Multiple indicators are used to evaluate the overall performance of the model. These criteria include convergence speed, F1 score, accuracy, precision, recall, and computational cost. In the most notable numerical results, the proposed strategy surpasses both the Finite Element and the Lattice Boltzmann methods. The presented method enabled fast convergence and reduced processing costs. These results were: accuracy (0.92), precision (0.93), recall (0.91), and F1 score (0.92). The proposed method is generalizable and adaptable, and it can address a variety of heat transport simulation problems in porous media. Unlike CNNs, which can identify significant spatial patterns, LSTM cells can only see temporal dynamics. These two components are required to show heat transfer, which is a continually changing phenomenon. Modern technology enables more complex simulations. Processing expenses are lowered, and estimations are more accurate. These two discoveries were obtained through the inquiry and methodologies. Finally, the CNN-FVM-LSTM technique simulates heat transport using complicated computer models. Predicting unusually high temperatures in porous materials may improve the model's accuracy, computational efficiency, and flexibility.