An enhanced multivariable dynamic time-delay discrete grey forecasting model for predicting China's carbon emissions

Abstract

The importance for the accurate forecast of carbon emissions affected by many factors is gradually emerging. Carbon emissions usually lag behind the related factors, which cannot be dynamically reflected in the existing grey forecasting models. Therefore, investigating the dynamic lag relationships remains the key challenge to carbon emissions forecast. For this purpose, an enhanced dynamic time-delay discrete grey forecasting model, denoted as DTDGM(1,N,τ), is proposed to predict the systems having dynamic time-lag effects. More specifically, a time-lag driving term consisting of both the interval and intensity of the time lags is developed to reflect the lag process of different factors to carbon emissions. The impulse response analysis of the vector autoregressive (VAR) model is carried out for determining the dynamic lags between carbon emissions and the related factors. In addition, a linear correction term is designed in the proposed model to extend the grey forecasting theory. Extensive experimental results about carbon emissions prediction from 1995 to 2017 show that the DTDGM(1,N,τ) model considering the delayed relationships can significantly improve the fitting and prediction performance of the model in comparison with the six benchmark models, including the three existing grey forecasting models, two machine learning models and one statistical prediction approach.