Applying BP neural network to detect conveyor belt fire with multi-sensors

摘要

A kind of forward neural network with three layers was applied to detect conveyor belt fire faster. And backward propagation (BP) algorithm was used to train the network parameters. The appropriate parameters and architecture of network were obtained training with 81 pair of data. Matlab was used to simulate and the experiment result shows training time is and error reduces most rapidly when ten neurons in hidden layer and momentum coefficient is equal to 0,95. Temperature, rate of temperature change, of carbon monoxide and rate of carbon monoxide dense change were considered as four parameters to detect the PVC belt fire in this paper. It is indicated that the network can give alarm as fire takes place about 350 s. The network can effectively detect the fire at the early stage of conveyor belt fire. At the same time, the reliability of alarm can be increased and the anti-interference capability can be enhanced when using this network.