Abstract: In order to improve the accuracy of fire alarm in a confined space, a set of techniques for the reliability of fire detection signal in a confined space was designed and established. The fire detection signals in a confined space whether meet the Gaussian distribution was verified by the experimental study of combustible smoke concentration with the neural network algorithm. The results show that the concentration of combustible smoke can be fitted into a quadratic Gaussian distribution within the allowable error range. The signal of CO for combustion obeys the Gaussian distribution and predicted value of aviation kerosene is more stable and has a smaller error value by algorithm than heptane, compared with that whether the CO concentration produced by the combustion meets the Gaussian distribution.

Key words: confined space, fire alarm, Gaussian distribution;neural network