The Grey Prediction Models And Application Considering The Data Characteristics

Since the establishment of grey system theory by Professor Deng Julong in 1982,grey prediction theory has become a system of prediction methods including GM(1,1)model,grey Verhulst and GM(r,h)models and related derivative models after forty years of development.These models have also been successfully applied to many scientific fields.However,with the further expansion of the application scenarios of grey prediction models,the time-series data are also more complex due to the combined influence of multiple factors.How to construct the corresponding grey prediction model according to the data characteristics has become a key problem to be solved.This paper summarizes the research results related to grey forecasting at home and abroad,and finds the problems faced by existing research and the entry point for new research.First,the fixed structure of existing models limits the adaptability of models to complex data characteristics.Second,the existing models fail to fully explore the nonlinear development trend and seasonal fluctuation characteristics of data series.In view of the above problems,based on the idea of “data characteristics driven” modeling,this paper constructs two kinds of models to fully tap the nonlinear and seasonal fluctuation characteristics of data sequences.The main findings of this paper are as follows.(1)Aiming at the data sequence with obvious nonlinear and irregular long-term development rate,a power-driven non-homogeneous discrete grey model is constructed.It solves the significant nonlinear problem that existing models are difficult to accurately describe the long-term development trend of renewable energy demand.Based on the idea of direct modeling and dynamic grey action,this paper constructs a power-driven nonhomogeneous discrete grey model,and finally realizes the effective capture of the long-term development trend of nonlinear data sequence.The prediction results meet the requirements of first-order prediction accuracy.Compared with existing models,the new model not only reflects the time-varying characteristics of external disturbance through dynamic time-varying parameters,but also uses the direct modeling method to avoid the potential information loss caused by accumulation and subtraction.The new model is more practical and has higher precision.Finally,the new model is applied to the prediction of annual demand for renewable energy in the three major economies,and the analysis is given according to the prediction results.(2)Aiming at the data sequence with significant nonlinearity and affected by seasonal fluctuations,a grey hybrid model based on STL-HP quadratic decomposition is constructed.It solves the problem that the fixed structure of existing models cannot handle the seasonal fluctuations and long-term trends of renewable energy systems due to the interaction of multiple factors.Based on the idea of “decomposition and integration”,this paper combines STL decomposition method with HP filter,and proposes a quadratic decomposition method based on STL-HP.Then,according to the characteristics of each component,the components are predicted and integrated by a more suitable long-term trend prediction model DPDGM(2,1)and a seasonal prediction model Holt-Winters.Finally,the grey hybrid model based on STL-HP quadratic decomposition is constructed.Compared with existing models,the data pre-processing method based on quadratic decomposition can achieve effective separation of trend and seasonal components of data series,thus reducing the difficulty of model construction.It also improves the accuracy of the model by making more accurate predictions of the components separately according to the data characteristics.In this paper,the new model is applied to the forecasting of monthly data of renewable energy demand and is compared and analyzed with other models.The results show that the new model can meet the demand of first-order accuracy for the prediction of data series with both long-term trends and significant seasonal fluctuation.Finally,the new model is applied to the forecast analysis of monthly demand for renewable energy from 2020-2022,and corresponding policy recommendations are made with a view to informing its continued development.