Research On Temperature Field Prediction Model Of Utility Tunnel Cable Fire Based On Support Vector Machine

All parts of the country are vigorously promoting the construction of urban underground utility tunnel,which contains a large number of power cables and gas pipelines with fire hazards.The fire problem in utility tunnel cannot be ignored.The spatial structure of underground utility tunnel is narrow and closed,which makes it difficult to obtain the internal temperature of the fire.Fire rescue workers lack real-time and effective fire information to infer the fire situation,which increases the difficulty of fire rescue in utility tunnel.Most of the existing fire temperature field prediction methods are limited by the experimental environment and conditions,which adaptation range is small and lack of real time performance.Therefore,it is necessary to develop an accurate real-time prediction model of fire temperature for fire prevention design and fire fighting in utility tunnel.The application of advanced means,such as big data and machine learning,has led to the rapid development in the field of smart fire fighting.Strengthening the research on real-time fire temperature prediction combined with machine learning will be beneficial to the safety and long-term development of utility tunnel.In this paper,a study on the prediction of fire temperature field in urban underground utility tunnel based on support vector regression algorithm is carried out.Based on n-heptane combustion experiments,a fire temperature field prediction model was established and applied to the cable fire scene in utility tunnel.The prediction performance of the model is evaluated through various aspects such as error analysis,evaluation index analysis,empirical formula comparison and running time.The main contents include:First of all,n-heptane combustion experiments were carried out in the laboratory to collect flame images and calculate its flame height.Then,a real-time prediction model of heat release rate based on the continuous variation characteristics of flame height is proposed using support vector machines.The prediction error of the model is ±10%,and the prediction time of a single sample is basically within 0.005 s.The results show that with the input of continuous flame height data within 1 s,the trained model can accurately predict the heat release rate of the flame at the corresponding moment in real time.Secondly,flame images and fire temperature data were collected during n-heptane combustion experiments in the utility tunnel.The flame heat release rate was obtained using the established heat release rate prediction model.A database containing parameters such as heat release rate,fire development time and thermocouple spatial location and temperature is constructed.Based on the database,a prediction model for the fire temperature field of the utility tunnel was trained and optimized.The error analysis shows that the relative error of model prediction for vertical and longitudinal temperature is±2%.Finally,a numerical simulation model of a standard section(200 m)of utility tunnel was established,and the characteristic fire parameters of cable fire were obtained by simulation.The established fire temperature field real-time prediction model is applied to the cable fire scenario to evaluate the practical ability of the model.Aiming at the weak prediction performance of machine learning methods around the fire source,a modeling optimization scheme is proposed to improve this phenomenon.Considering the prediction performance and running time of the model comprehensively,the model was found to be optimal for a thermocouple arrangement spacing of 5 m near the fire source.