关键词: 自然发火期, 主成分分析, 神经网络, 预测

Abstract: The coal self-ignition duration is an important parameter not only to measure the characteristic of coal self-ignition but also to direct the fire control technique in coal mines.The BP neural network prediction model for the self-ignition duration is established based on the Principal Component Analysis（PCA）.Inducting PCA to pre-analyze the original multi-objective variables，and using the principal components of original variables as the input of network can cut down the dimensions of input，and at the same time eliminates the relativity between variables，so improves the convergence speed and stability of network and simplifies network structure.Testing actual instances to validate that the PCA-BP neural network compared with the normal neural network improves the precision，reduces training time and possesses better performance.

Key words: self-ignition duration, Principal Component Analysis（PCA）, neural network, prediction