Research Of Urban Fire Risk Assessment Based On Fuzz Information Optimization Method

Fire risk assessment is an important component of fire science and engineering. And the research of relationship between urban fire risk and socioeconomic factors becomes a research focus. There are so many indexes which can be used to describe urban fire risk, but each of them only characterize fire risk from one aspect, still with redundant information among them. In this paper, four common factors, which are the fire total damage factor, the deaths on average factor, the loss on average factor, and the fire probability factor, have been achieved from 13 frequently-used fire statistical indicators by method of Factor Analysis. These integrated factors are found to be meaningful and independent. Furthermore, the relationships between the common factors and a number of socio-economic factors are analyzed by the method of multiple linear regressions. Based on the data using in this paper, it can be concluded that the fire total damage factor shows a significant linear correlation with the population; the fire probability factor is mainly influenced by three factors including the proportion of children, the population mobility and the living conditions, and has no linear correlation between the population; the economic level and the educational level; the deaths on average factor would be reduced according to the increase of the fire training level; and there is no linear correlation between the loss on average factor and any one socio-economic factors considered in this paper.When analyzing small sample in used of frequency histogram, the shape and trend of histogram are sensitive to the first statistical point, interval length and singular data. The numbers of great fires in Japan between 1995 and 2008 was analyzed based on One-dimensional information diffusion. The results showed that information diffusion is more stable than frequency histogram in dealing with small sample problem. The probability of controlling points can be fitted with the Gaussian distribution. Based on information diffusion, a method of great fire risk prediction is raised up; exceeding probability of the number of great fire annual in Japan was calculated. Great fire risk in Japan was predicted reasonable.After analyzing the relationship between fire incidence and single socio-economic factor in used of different methods, it is concluded that fuzz information approximate reasoning method is better than linear regression in dealing with fuzz information. Adding another socio-economic factor, redid the identification, firstly the dependent variable is of better interpretability due to the increasing of independent variable number; the result proved again that information-diffusion approximate reasoning is able to digging more information form fuzz sample than linear regression, its processing results make more on the interpretation of the dependent variable. Training the sample data by normal BP network, it is found that BP network convergent quickly, but with low prediction accuracy. Based on Hybrid Model, network convergent more quickly, and its prediction accuracy is higher than that of BP network. It's proved that information-diffusion approximate reasoning is effectively in smoothing the sample, and can convert incompatible mode to compatible mode. It also has good prediction precision.