Study On Multi-sensor Composite Fire Detection Information Fusion Alarm Method

Application of fire detectors is a very important means to achieve fire detection.More and more types of fire detector have been developed and widely used after years of development.However,such problems of sensor false alarm and nuisance alarm come along,which seriously affects the credibility of detector alarm.In order to solve such problems,multi-sensor detectors were developed and optimized based on two or three kinds of sensor.So how to deal with multi-sensor information problem(collected information characteristic,data redundancy,the characteristics of multi-sensor fire detection,portfolio selection and multi-sensor location distribution),and the information fusion decision problem become the new research subject.In this paper,a model of fire detection feature combination selection in multi-sensor fire detection is proposed.With the intention of optimization of multi-sensor fire detection sensors and selecting the combination of high performance characteristics,based on the NIST(National Institute of Standards and Technology)studies of fire alarm response experiment as a benchmark test,some of these experimental data were selected to optimize multi-sensor fire detection sensor number,combined with the concept of mutual information in information entropy theory,Correlation between different characteristic combination of fire and fire state,and the redundancy of information with different characteristic combination were analyzed.Thus,efficient feature combination is selected.Then,combined with BP neural network and genetic algorithm,genetic algorithm was used to optimize the initial structure of BP neural network,and a combined model based on genetic algorithm was proposed to optimize the BP neural network--GA-BP neural network model.The global search function of genetic algorithm is combined with the prediction ability of BP neural network to make the final prediction result more accurate.Finally,some experimental data of SH1(standard smoldering fire),SH4(standard open fire)and typical interference sources were selected as training samples,and MATLAB software was used for simulation to verify the feasibility and effectiveness of this fire detection method based on information fusion technology.