The Research Of Fire Detection Based On Wavelet Neural Network Algorithm

With the improvement of the desire of fire detection system today, many new fire detection methods have been used for fire detection. The fire detection systems based on intelligent information processing have been proposed to handle this situation. These systems have self-learning and self-adaptive. The study of them has been the direction in the study of fire detection technology.The fire parameter gathered by the detectors is unable to know in advance, the non-constitutive signal. The traditional survey methods merely carry on the judgment and the recognition through gathering sole fire characteristic parameter information. So it is disturbed by the environment inevitably. The question about the system's high mistakenly reporting rate is prominent. In recent years, the accuracy of fire alarm becomes better, while the sensibility and reliability of detectors are improved. But it can't match the request of the automation of the fire detection system. It can decrease mistakenly reporting only by describing fire inherent characteristic completely and exactly. Thus fire detection algorithm based on multi-criteria is the main research direction in the present fire surveying field.Through researching the fire mechanism and the present fire surveying methods, and combined with fire detection system's own characteristic, this article proposes fire detection algorithm based on wavelet neural network algorithm.In this paper, a gas sensor array was used to collect signals. Because wavelet neural network has the following merits:high precision, learning rate fast etc, we use wavelet neural network in the field of inspection of these fire gases.In the basis of one dimension wavelet neural network, we researched two different structures of wavelet neural networks. And we used them into the inspection of these polluted gases. We used them because the constringent speed of wavelet neural network is very fast, they are not sensitive to the inputs of the network and they have the characteristic that they also can effectively approach the functions or signals.Classical neural networks mostly train the network with back propagation algorithm. But the back propagation algorithm often gets into the minimum value and its constringent speed is slow. Aim at the gases, we design a optimum back propagation algorithm. We improved the Classical back propagation algorithm on two points:one is that we adopt a method of self-adaptive learning rate algorithm that is based on the sign change of the grads, second is that we adopt momentum item. Finally we contrast the result of the experiment, and find that the optimum algorithm can effectively improve the training speed of wavelet neural network and avoid that algorithm get into minimum values. The experiment proved that we can well deal with the gas signals with wavelet neural network. The inspection of these polluted gases result is satisfying.