Multi-scale patch transformer with adaptive decomposition for carbon emissions forecasting

Rapid urbanization and industrialization have led to a significant increase in carbon emissions, posing a challenge for sustainable environmental management. However, current research predominantly focuses on traditional forecasting models that often overlook the complexity and dynamic nature of environmental data. To address this, a novel multi-scale patch transformer with adaptive decomposition (MPDformer) has been developed specifically for forecasting carbon emissions. This model introduces an adaptive decomposition method that dynamically assesses the noise level, trend, and stationarity of data to select the most appropriate decomposition technique. In addition, the use of a Transformer with multi-scale patches can optimize the use of information at different granularities in the decomposed sub-series for time series prediction of carbon emissions. Experimental validations have shown that this method possesses an exceptional capability to discern complex temporal dependencies within fluctuating environmental data, consistently outperforming comparative models across a range of carbon emissions datasets and various forecasting horizons. These results indicate the potential for more accurate and reliable carbon emissions forecasts, which can contribute to better-informed decisions in sustainable energy planning and environmental management.