Abstract: In order to solve the decision inference problem and the information redundancy problem in double-layer distributed decision in big data environment, a hypothesis that there exits correlation in inference information is made based on the analysis of the complexity of the participating subjects and the overlapping of business relations in such decision, and an inference information description mechanism with the advantage of soft expression is suggested by employing the BPA function in evidence theory. On the ground of that, theorems and inferences are constructed for upper and lower departments to eliminate the influences of both administrators and upper departments and used to make a scientific fusion of all the decision subjects' meta-inference information. Finally, the procedures of double-layer distributed decision in big data environment are constructed according to the "upper to lower" decision order. The result of numerical comparison analysis shows the present method is scientific and feasible. The present method is benefit to develop the thought for solving management decision problems in big data environment and explore the "big" mode for dealing with incompleteness data or correlation data.

Key words: big data environment, double-layer decision, distributed decision, evidence theory, information fusion