一种基于小样本数据信息扩散的重大火灾频度估算方法

重大火灾的发生是小概率事件,其历史数据属于小样本统计数据。在应用频率直方图法对小样本数据进行处理时,分析结果对直方图的起点、区间步长以及样本中的奇异数据十分敏感。采用信息扩散方法,对日本1995-2008年各年重大火灾次数统计样本进行分析,对比了不同扩散条件下的扩散结果,结果表明信息扩散方法具有很好的稳定性和一致性。基于信息扩散的结果,计算了以年为周期的重大火灾发生次数的超越概率分布,建立了一种重大火灾频度估算方法。

A Method of Serious Fire Risk Prediction Based on Small Sample Information Diffusion

The serious and extremely serious fires are events with low probabilities, and the statistics data of them are of small sample. The common frequency histogram method is not applicable to analyze the small sample data, because it is sensitive to the analysis parameters such as the starting point and the interval step. In this paper, based on the method of information diffu- sion, the statistics data of the serious fires in Japan between 1995 and 2008 are analyzed. The results show that the information diffusion method has good stability and consistency. It has been proved that the influences of changing the starting point, the interval step, or the diffusion function can be ignored. Based on the results of information diffusion, the probability of the num： her of the serious fires per annum can be fitted with the Gaussian distribution.