The Algorithm Research Of Fire Detection System Based On The Multi-Source Information Fusion Method

The intensive development of the high-rise buildings of the city will not only increase the hidden dangers of fire, but also make the fire detection and alarm more difficulty. Because the fire signal is uncertain and susceptible, so if we adopt current widely used method based on the detection of single characteristic signal to detect fire, it will lead to a high rate of wrong or missing alarm. Moreover, The fire detection system cannot effectively resist the interference and adjust to the all kinds of changes from environment.In order to get rid of the defect of the traditional fire detection method, this thesis puts forward an intelligent fire detection system based on the method of multi-source information fusion, after expounding the basic concepts and theories of the multi-source information fusion. With the requirements of multi-source information fusion, the system takes various types of fire characteristic single into consideration, such as the potency of CO and smoke and the change rate of temperature, in order to overcome the defect that the one-sided description of extern researched object from single fire characteristic signal. According to the hierarchy of multi-source fusion system, the selected fire characteristic signals are treated on those three layers such as information layer, feature layer and decision layer. In order to reflect the intelligence of the system, two artificial intelligence methods such as artificial neural network and fuzzy reasoning technology are used to process fire signal on the feature layer and decision layer. Both two AI methods can make the fire detection system possess those characteristics such as self-learning, self-adapting and intelligent reasoning. They can also improve the reliability of the system to identify fire. Especially, on the feature layer of the system, this paper proposed RBF-BP mixed neural network by combining pure BP and RBF network with each other. The mixed network can make the advantages and disadvantages of two single networks complementary and get superior network performance. Applied into the fire detection technology, the mixed network can effectively improve the accuracy of recognition of fire.Through using the fire detection simulation software to train sample data, different performances of three different kinds of neural networks are compared to confirm that the mixed are more superior than the other two pure networks. Furthermore, the software can also verify that it is feasible to apply the idea of multi-source information fusion, artificial neural network and fuzzy reasoning technology into fire detection system to improve recognition performance.