A Research Regarding Detection Method Of Aircraft Cargo Compartment Fire Based On Recurrent Neural Network

It is of great significance to reduce the false alarm rate of aircraft fire and realize fast fire identification for flight safety.At present,fire alarm incident in the aircraft cargo is rather complex,such as passenger luggage burning,aircraft electrical equipment aging,dust,water vapor and other particulate matter caused by false alarms.When there is a real fire,there is often multiple information generated,such as temperature changes,smoke concentration changes and many types of gas generation.For one thing,different types of fire lead to different products produced during combustion.For another thing,the variation of characteristic parameters of fire source will also change at different stages with the passage of time.Therefore,considering the above problems,a fire detection method will be proposed,which combined with the time dynamic information,thus the fire characteristic parameters are fused to make a comprehensive judgment.Then it will get a more complete information of what kind of fire is real or fake.First of all,considering the characteristics of fire combustion,a new type of fire detector has been developed,which can realize the collection of smoke concentration,temperature,CO,CO2,SO2 and other information and then use the upper computer software to display real-time data.Secondly,according to the existing deep learning technology,a fire detection algorithm has been proposed based on LSTM cyclic neural network,which tried to classify and identify the truth of fires by using the time series parameter data.Finally,several groups of real and fake fire source experiments were carried out in the simulated cargo hold of the aircraft,whose data were analyzed and processed.Then the data were connected by time to form feature sequences to train and verify the classification model of fire detection algorithm.Through different selection methods of time series length of data set and the intensity of simulated burning process,the robustness,rapidity and accuracy of the fire detection algorithm are verified.The results show that the fire detection algorithm based on LSTM-RNN neural network can effectively learn the long-term dependence of time series data.The network has achieved good results in the fast identification of real ignition source and interference source.For data set of 3 seconds the identification accuracy of true ignition source and interference source can reach 88% and 97%,respectively.