Research, Early Fire Detection Method Based On The Process Of Feature Information

With the advance of human civilization and the rapid development of science and technology,fires occur frequently and become one of the greatest threats to people's lives , property and social security . Therefore, it is great important to research the mechanism of fires in achieving accurate and reliable early detection and warning.Most of existing methods of fire detection are analyzed in the thesis, it is indicated that gas detection that use CO/CO₂ as its characteristic target can be more effective and reliable in early fire detection comparing to conventional means.Researchers use a specific experiment system for early fire warning research based on FTIR technology to make a lot of experiments to simulate fires of limited space. The spectrums of CO/CO₂ are collected and the concentration is got by spectral analysis.Fires takes on a character of non-stationary dynamic process, fires are often identified only by transient data according to conventional detection methods. It is so unreasonable that may cause misreporting. Therefore this thesis brings forwards an early fire detetion based on process-characteristic information. With the utilization of window function and least square method, we distill the process-characteristic parameters of CO/CO2, with the rejection of random element. The parameters include the concentration, the rate-of-rise and the acceleration-of-rise of the characteristic gas, is the representation of the process-characteristic information of the fire. We focus on the study of changing law of the process-characteristic parameters and discover the distinctions of flaming fires, smoldering fires and nuisance fires.Based on the study of the theoretics and the practice of experiments, we set up a RBF network with three inputs including of the parameters of CO rate-of-rise and acceleration-of-rise as well as CO2 rate-of-rise. The three outputs of the RBF network are the identifications of flaming fires, smoldering fires and nuisance fires. After the training and simulating of network, it is proved that the fire detection based on process-characristic information and RBF network can provide an accurate early alarm without misreporting. Comparing to conventional fire detector, this new method will provide more time for fire preventing.