Abstract: Compositional data are widely used in the field of society, economy and technology. This paper discusses two different forecast modeling methods of compositional data .The first method supplies a forecast model making use of logarithm change. Sphere projection, a non-linear approach for dimensionality deduction, is used in the second method to build a forecast model. With the contrastive study, the paper lists out the advantage and disadvantage of each. The methods talked above are respectively used in the compositional data of the Beijing employees from 1994 to 2000 that are divided by proprietorship. In the end of this paper, the status of employment in Beijing is analyzed.