In order to predict GDP change and trend in a period of future, this paper proposes time series combination forecast model based on the combination of information granulation algorithm and support vector regression machine algorithm, makes fuzzy granulation processing on the GDP values from the first quarter of 1994 to the forth quarter of 2013, uses crossover experiment training to seek the optimal parameter of support vector regression machine and conducts regression forecast based on three fuzzy parameters such as Low, R and Up and obtains GDP change and trend in each quarter of 2014 which is identical to the practice. The results show that this algorithm has good generalization and can be used as prediction model for time series data. 