【摘要】 为监测建筑火灾事故区域的危险程度，实现更加安全、高效的火灾应急救援，以通廊式建筑为研究对象，基于转置卷积神经网络及数值模拟方法开发1种可实时预测走廊位置处烟气扩散和温度分布的神经网络模型。首先，依托Python建立包含全连接、转置卷积、反池化等在内的19层神经网络模型的整体架构；其次，建立包含99个火灾场景，共7 920组图像数据的火场信息数据库用于模型训练；最后，使用测试集对模型进行可靠性验证。研究结果表明：烟气(温度)预测模型在不同火灾场景下的预测精度达到95%,训练完成后模型的预测时间一般为1～2 s。研究结果可为应急策略的快速制定提供数据参考。更多还原

【Abstract】 In order to monitor the degree of danger in the area of building fire accidents and realize the safer and more efficient fire emergency rescue, taking the gallery building as the research object, a neural network model for the real-time prediction of smoke diffusion and temperature distribution at the corridor location was developed based on the transposed convolution neural network(TCNN) and numerical simulation method.Firstly, the overall architecture of a 19-layer neural network model including full connection, transposed convolution and inverse pooling was established based on Python.Secondly, a fire scene information database containing 99 fire scenarios and a total of 7 920 sets of image data was established for model training.Finally, the reliability of the model was verified by an independent test set.The results showed that the prediction accuracy of smoke prediction model and temperature prediction model reached 95% in different fire scenes, and the prediction time of the model after training was generally 1～2 seconds, which can provide data reference for the rapid formulation of emergency strategies.更多还原