Application of Convolutional Neural Networks and Long Short-Term Memory networks in real-time prediction of fire source intensity
To solve the problem of real-time and accurate identification of fire location and intensity in fire scenes,a real-time fire source intensity prediction model was built by integrating the algorithm advantages of Convolutional Neural Networks(CNN)and Long Short-Term Memory(LSTM)networks.The CNN is good at local special feature extraction,and the LSTM is good at time series capture.The model can accurately predict the fire location and intensity in fire scenes through the real-time sequence data received by the temperature sensor in the building.A Fire Dynamics Simulator(FDS)was used to recreate the fire scenes and collect real-time sequence data by temperature sensors.The temperature sequence data was analyzed and used to establish the fire scenes database.The prediction model was trained to mine the mapping relationship between fire source intensity and fire scene data,and evaluated for accuracy,timeliness,and robustness with a case.The results show that the convergence speed of fire location prediction is faster,compared with fire intensity prediction,fire location,and intensity prediction.Through short-length sample data,the trained model can accurately predict the fire location and intensity in fire scenes in real-time.The prediction accuracy is higher than 94.04％with more than 2-length of the sample.When the sample length is 10,the prediction accuracy of fire location is 99.39％,the prediction accuracy of fire intensity is 99.70％,and the prediction accuracy of both location and intensity is 99.18％.Moreover,the prediction accuracy of both location and intensity can still exceed 95.10％when the temperature sensor is damaged at intervals and the damage rate is less than 70％.If the temperature sensor is constantly damaged at one end or both ends of the corridor,the damage rate shouldn't be higher than 30％or 50％,respectively,to guarantee greater than 90％accuracy.The fire source intensity prediction model provides a reference for obtaining fire scene information in real-time.
NETL
NSTL
万方数据
维普
