Research On Technique Of Grey Prediction Model

The purpose and task of prediction is predict the thing which has not occurred. Regardless of the individual or the organization, prediction is an important part of the future-oriented decision-making process. It is also an important prerequisite for scientific dicision-making.Among the many predicton ways, grey model has been valued by many scholars since creation. It has many advantages, such as it needn't many sample and doesn't requst sample distributed regulation. It also has advantages of less computation and stronger adaptability. Grey model has been widely used in various fields and obtained glorious achievement . However, grey prediction model needs further improvement. The factors that affect the precision of the grey prediction model are analysed in this paper.Some valuable improvements are reached. The main work includes:Firstly,Data function conversation is an important method to improve prediction accuracy of grey model. Based on the analysis of data sequence smooth,a new function converse method is put forward. This function converse method is proved effectively by theory proof and experimental test.Secondly,Grey modeling mechanism is analyzed. It is also discussed the influence of initial value to the grey prediction accuracy.A new initial value selection method is proposed.This method can improve the accuracy of grey prediction model. Based on the characterictics of grey model and neural network, grey neural network forcasting method is put forward which have advantages of both grey model and neural network.Lastly,The building method of grey prediction model background value is analyzed. The Vector is introduced into the computing formula of background value array which change the original grey prediction model to construct the background value with a fixed value.Particle swarm optimization algorithm is adopted to solve the background value.This method is applied to grey nonlinear model GM(1,1,(?)b) .A new grey nonlinear model GM(1,1,λ,(?)b) is contructed which has wider application field and gives better precision than grey nonlinear model GM(1,1,(?)b).