The Research On Application Of DS Evidence Theory Model Based On BP Neuarl Network In Fire Detection

In China,more and more people enter into the city with the advancement of urbanization.A large number of people gather in the communities,office buildings,shopping malls and other small areas,which undoubtedly increase the threat of fire.So a more reliable fire alarming device is needed to protect people's lives and properties.The core of the fire alarming device is the fire detection system.It is a single sensor that collects fire signal in traditional fire detection.But with the environmental variation and social development,this method shows more and more limitations.There are more and more factors that interfere with the normal operation of the sensor.So a new fire detection method is needed to meet people's needs.Multi-sensor data fusion technology is widely used in many fields.In the fire detection,the use of multiple sets of sensors is able to make up for the shortcomings of traditional fire detection.DS(Dempster-Shafer)evidence theory and BP(Back Propagation)neural network are also widely used among the data fusion technologiesDS evidence theory has a good performance in dealing with problems of uncertainty.For the information collected by the multi-sensor,the DS evidence theory is gradually able to reduce the uncertainty of information in the fusion process,and finally get a satisfied fusion result.However,the determination of basic probability assignment in DS evidence theory is a problem.BP neural network has the advantage of adaptive self-learning,and its output can be controlled in the range of [0,1].So we combine the BP neural network with DS evidence theory,considering the output from the BP neural network as the basic probability assignment of DS evidence theory.In view of this idea,this paper establishes a DS evidence theory model based on BP neural network to realize the detection on fire.The final experimental result shows that this model has a good ability in identifying the fire,and has a anti-interference.The main work and achievements of this paper are as follows:(1)For the limitation of single sensor in the application of fire detection system,the DS evidence theory model based on BP neural network is proposed.The effectiveness of this model is verified by simulated fire experiment.The result shows that the model has a good ability in identifying the fire,and in the case of interference it is still able to accurately identify the fire.(2)The key to the construction of BP neural network is the determination of parameters.In this paper,the BP neural network is constructed by the knowledge and experience from a number of scholars and tests.In order to speed up the convergence of BP neural network,this paper uses L-M(Levenberg-Marquardt)algorithm which is an improved algorithm for the standard BP algorithm.The result shows that the L-M algorithm greatly accelerates the convergence of the network.(3)In the simulated fire,the high conflict evidences generated by the interference data cannot be fused with other evidences by the synthetic rule of DS evidence theory.To solve this problem,a new method of dealing with high conflict evidences is put forward.The method preprocesses the data,which reduces computational effort.As a result,the processing speed is fast.The result shows that the method can greatly reduce the conflict between the evidences and make the result more reliable.