Improvement of residual error in grey GM(1,1) model

When building model base on residual sequence, non-negative process on residual sequence should be considered if both positive and negative signs appear simultaneously. In this paper improvement of the conventional method of non-negative process is presented which can not only reduce the deviation of class ratio but also remain reductive error of both simulative residual sequence and simulative original sequence unchanged or be reduced. Finally, a numerical example demonstrates that the proposed model can achieve high precision simulation and forecast effect.