| The theory of the grey system based on cybernetics can excavate the evolution law of the uncertain system of "small sample,poor information",so it has been widely used in many fields such as energy,economy,and environment since it was put forward.With the development of application and research,scholars have improved the grey prediction model from different angles,to further strengthen the gray prediction model of the uncertainty system intrinsic information mining and grasp the law of system evolution.However,with the evolution of China’s energy revolution,the existing grey prediction models are often faced with complex time series with high complexity,uncertainty,and sparsity.And the current grey prediction model is difficult to mine the different characteristics of the real-time series to meet the different needs of energy prediction.Given this,to explore the development characteristics and the evolution law of the system and provide a reference for national energy planning and energy strategy adjustment,this study,based on the study on the mechanism of the technology of grey action quantity optimization in grey prediction theory to optimize the grey prediction model,constructs three kinds of new grey prediction models from the perspective of knowledge-driven and data-driven based on the grey generation technology,grey action quantity optimization technology,and discretization technology in grey prediction theory to solve the three types of problems widely existing in the existing grey prediction model.And the study also puts forward a kind of data-driven adaptive selection algorithm of the structure of the grey forecasting model.The proposed new model not only optimizes the grey accumulation generation technology and improves the feature extraction method of sparse data,but also optimizes the structure of the grey prediction model,expands the application scope of the grey prediction model,improves the mining and representation ability of grey prediction model for complex system evolution mode,and promotes the further development and application of grey system theory.Based on the analysis of the mechanism of grey action quantity optimization technology and discretization technology,this study has done the following three points:First of all,to effectively identify the seasonal characteristics and trend characteristics of seasonal time series in different periods for the short-term prediction demand of seasonal time series,a seasonal fractional-order discrete grey prediction model based on moving average filter is proposed.Based on the moving average filter,the new model realizes the identification of seasonal characteristics and trend characteristics of seasonal time series.Based on the PSO algorithm,the co-optimization of trend and seasonal characteristics is considered based on fractional-order and initial value optimization.Besides,since the improvement of the new grey model is based on the discrete grey prediction model,the structural error and information distortion existing in the traditional grey prediction model are effectively avoided.Secondly,a seasonal discrete grey prediction model based on periodic aggregation generation operator is proposed to meet the long-term forecasting needs of seasonal time series,which solves the problem that the traditional grey forecasting model cannot fully adapt to the nonlinear evolution law of the system and the error inceases exponential with the growth of the prediction period.Based on the periodic aggregation generation operator and dummy variables,the new grey model improves the adaptability of the model to the periodicity,realizes the compatibility of the model for medium and long-term prediction and short-term prediction,meets the differentiated prediction needs of different prediction subjects,and improves the accuracy of long-term prediction.Besides,the new grey forecasting model replaces the traditional grey action quantity with dynamic grey action quantity,which enhances the adaptive ability of the model to the nonlinear development trend of time series.Thirdly,for the above two types of grey prediction models,the representation of seasonality of time series is only reflected by seasonal factors,which simplifies the differentiation characteristics of complex time series.The mining of time series characteristics is driven by expert knowledge,and the potential frequency domain characteristics of complex time series are ignored.Thus,from the data-driven point of view,based on the grey action quantity optimization technology and discretization technology,this paper further proposes a structural adaptive discrete grey prediction model based on grey action quantity optimization.Through the introduction of nonlinear term and periodic term,the new grey forecasting model not only strengthens the ability of the traditional GM(1,1)model to capture the nonlinear and linear development trend of time series,but also improves the adaptability of the grey prediction model to any periodic term time series from the structure of grey prediction model,and improves the characterization ability of grey prediction model to the characteristics of the system.Besides,because of the high degree of freedom of the new grey model structure,it is easy to produce the problem of over-fitting or under-fitting.Given it,a data-driven model structure adaptive selection algorithm is proposed,which improves the mining ability of the model to the potential characteristics of the system.This method can effectively enhance the accuracy of the prediction and reduces the demand for the knowledge of the modeler. |
