Research On Optimized Grey Prediction Models And Their Applications

Grey forecasting theory,one of the main methodologies solving uncertain systematic forecasting problems,has gained extensive attention of global researchers ever since it was built.With the constantly developing of this theory,it has been applied to many economic and social fields.However,dute to the complexity and nolvelity of the emerging problems,which the existing grey forecasting models can not handle with,it is necessary to optimize and modify the traditional grey forecasting models for satisfy the pratical application requirements.Therefore,main works that optimize the GM?1,1?,grey power models,GM?1,N?and the discrete multivariable grey models are as followed:?1?Optimizing the equally-spaced and unequally-spaced GM?1,1?model by ultimizing the modified initial condition that is calculated by using weighted sum method.For the equally-spaced sequences,this thesis uses the weighted value of each component of 1-AGO sequence as the initial condition based on the principle of new information priority.By using this new initial condition,the novel NOGM?1,1?model is designed.Subsequently,the intelligent algorithm is employed to search the optimal weight parameter and time parameter under the principle of minimun average relative error.As for the unequally-spaced series,the properties and the effect on the modelling precison are studied under the contractive transformation and optimization of the initial point.Additionally,under the principle of minimizing the square sum of the relative error between the original series and the forecasted sequences,the two optimized approaches that modifying the time response function and initial point respectively can obtain the similar predicting accuracy.Subsequently,a new unequally-spaced NIUGM?1,1?model is proposed,which has the novel initial condition that is optimized by using the weighted function under the principle of new information priority.?2?Aiming to solving modelling problems of the nonlinear system having sparce data,two optimized models,namely NIGPM?1,1?and MDGPM?1,N?,are built.For the univariate nonlinear problems,the NIGPM?1,1?model is put forward,which has a new initial condition that is designed by using weighted sum method considering the diverse weights between the new and old information.Subsequently,by combing this newly proposed initial condition.the power index in this new model can obtain its optimal estimate value.To deal with the multivariable nonlinear problems,a new multivariable discrete grey power model,DGPM?1,N?,is initially proposed.Considering the great effect of driving factors on the modelling precison,a novel modified model,MDGPM?1,N?,is built by introducing a driver control function to recognize the main driving factors in driving stages.Subsequently,approaches to identifying the parameters of the driver control function are discussed on the premise of having sufficient information,and the modeling steps are summarized.?3?Based on the interactive relationship among driving factors and the data characteristics of the driving factors,The IEGM?1,N?and DVCGM?1,N?models are put forward,resectively.In order to address the problems caused by the interactive relationships among the driving factors,a new model,namely IEGM?1,N?,is designed by introducing the interactive item into the GM?1,N?model after sdudying the concept of interaction effects.Subsequently,the methods of estimating the parameters are put forward,and two derived models are proposed to extend the application fields of the novel model.In addition,the application scopes of the extended models are further studied,and they are proved to be equal to the original model.To solve the problems that the influencing factors have dummy variables,such as policies and gender,the DVCGM?1,N?model is designed by introducing the dummy and quantitative variables into the GM?1,N?model.Additionally,due to the great effect of the background value on precison,the PSO?Particle Swarm Optimization?method is employed to seach the optimal parameter in the background value.Subsequently,the optimized DVCGM?1,N?model having optimized dummy variables and background values are put forward to extend the application areas of the grey system theory.?4?Considering the accumulative effects of the influencing factors on the system variables,a multivariable time-delayed discrete grey model,namely TDDGM?1,N?model,is proposed.Due to the complexity and uncertainty of the time-delayed mechanism of the driving items,the deficiency of the traditional multivariable grey model is analyzed and the TDDGM?1,N?model is proposed by introducing the time-delayed coefficient.Then,the approaches of calculating the parameters are discussed.According to the sufficient and insufficient information concerning the time-delayed coefficient,the empirical analysis and PSO algorithms are used to search the optimal estimation of the time-delayed coefficient,and the modelling procedure is listed.?5?Forecasting the CO₂ emissions in China by comprehensively using the above optimized grey models.Firstly,the characteristics of the CO₂ emissions are analyzed basd on their current situations.Subsequently,the influencing factors of the CO₂ emissions are recognized by summarizing the previous studies.Thirdly,the development trends of the CO₂ emissions and their influencing factors are estimated by using the DGPM?1,N?and the NOGM?1,1?model,respectively.Finally,according to the forecasted results,some suggestions on controlling China's CO₂ emissions is proposed.