Abstract:

To address the complexity of fire environment and the difficulty in predicting the temperature field in modern commercial buildings, a fire temperature field prediction model was constructed by combining CNN with SVM. Firstly, FDS was used to construct a commercial building fire model, and the sequence data received by the temperature measurement points were obtained. The temperature, position coordinates, and fire duration were used as input parameters to build the dataset. Secondly, the Rime Optimization Algorithm (RIME) was introduced to optimize the number of hidden layer nodes, regularization coefficient, and learning rate in the CNN-SVM, and then the prediction model was established. Finally, experiments were conducted based on the established dataset and prediction model, and the anti-interference ability of the model under different sensor failure rates was discussed. The results show that the model performs optimally in the prediction of the temperature field plane, with an average absolute percentage error of 5.6% and a maximum relative temperature error not exceeding 25%. The anti-interference performance is the best under three working conditions, and the maximum error does not exceed 15% under extreme conditions.

Key words: commercial building fires, temperature field, fire dynamics simulator(FDS), convolutional neural networks (CNN), support vector machine (SVM)