Abstract: This paper deal with the dynamic model and prediction method under the condition of instable sequence, and the original data of forest fire from 1950-1984 of Guangxi were used. The auther combined the time-vary, self regression model with stimulus function to analogue the occurence of forest fire; the co-and low-boundary envelop curve model were used to analogue the variation range of diastor sequence and the grey topological model to the fire occuring value that the years had happened. By the fitting check and forecaeting to the degree, range, diaster frequency of this province in 1985-1987, the good result of fitting and forecasting was obtained and this model and method provided by the author also proved practically suitable.