Research On Temperature And Visibility Prediction Of Tunnel Fire Based On Neural Network

In the event of a tunnel fire,its rescue is difficult and hazardous.Real-time prediction of tunnel fire characteristics can provide data support for fire rescue and reduce casualties.Research on tunnel fire characteristics has been relatively complete,but mostly carried out through experiments and simulations,but there is less research on real-time prediction of tunnel fire temperature,visibility and other characteristics.Thanks to the development of computers,the real-time prediction function of neural networks is widely used in various industries.Therefore,this paper simulate the changes of roof temperature and visibility at 2 meters height in tunnel under different fire source location,Heat Release Rate and ventilation wind speed,and combines LSTM and FCNN neural network to establish tunnel temperature prediction model and visibility prediction model to achieve real-time accurate prediction of temperature and visibility along the underwater tunnel,which provides technical support for tunnel emergency.This will provide technical support for emergency rescue and evacuation of people in underwater tunnels.The main work and results are as follows.(1)A numerical model of underwater tunnel fire is built based on an actual rivercrossing shield section tunnel in an engineering project,and the prediction data sets of fire temperature and visibility characteristics are established.A total of 100 sets of fire conditions with 5 fire source locations,4 Heat Release Rate and 5 wind speeds were simulated by FDS software to collect temperature and visibility data and build the temperature and visibility datasets for training the neural network prediction model.(2)Based on LSTM and FCNN neural networks with temporal memory for data sequences,the temperature prediction model and visibility prediction model were developed and optimized to achieve accurate and fast prediction of temperature and visibility along the tunnel under fire conditions in the dataset,with high model prediction efficiency and millisecond response time.The specific conclusions are as follows.1)The optimization of temperature and visibility prediction models was achieved by studying the effects of different building parameters on the models,and the average prediction accuracies of LSTM and FCNN temperature models were 97.81%and 94.66%,respectively.the prediction accuracies of LSTM and FCNN visibility models were 98.84% and 98.13%,respectively.2)The prediction results of the temperature and visibility models were analyzed in terms of accuracy,training time,and scalability,respectively.Both prediction models can predict the temperature and visibility under a fire condition in millisecond time range.In terms of scaling performance,the prediction accuracy of the models becomes progressively smaller as the gap between the validation conditions and the fire parameters of the conditions in the dataset increases.3)Evaluation of the application of temperature prediction models and visibility prediction models in tunnels.Both LSTM and FCNN models can accurately and quickly predict the changes in temperature and visibility along the tunnel for fire conditions within the dataset;in fire conditions outside the dataset,the temperature model is more accurate in predicting temperature values downstream and upstream of the fire source with small temperature changes,and the visibility model is more correct for areas far from the fire source than for areas near the fire source.