Abstract: To further broaden the application of neural network algorithms in the field of fire detection, a national standard fire experimental data with a smoldering fire, open fire, and no fire occurrence probability as output results was used to train a multi-layer BP neural network on the MATLAB platform. The decisive coefficients of the network for the sample fitting of the test set are more than 0.95. Using AT89C52 as the main control chip, a simulation circuit diagram which can meet the purpose of fire detection is designed in Proteus. By introducing the trained BP neural network parameters into the simulation formula of the neural network, the software program that can drive the circuit action is designed, and the circuit simulation test of the fire detection system is realized.

Key words: fire detection, 51 single chip microcomputer, MATLAB, BP neural network