Carbon emissions forecasting based on tensor decomposition with multi-source data fusion

Abstract

Accurately forecasting carbon dioxide emissions is crucial for policymakers and researchers aiming to combat climate change and develop effective emission reduction strategies. This study introduces an innovative method that leverages multi-source social media information to address the challenges of insufficient information and data uncertainty in carbon emission time series forecasting. We propose a combined Tensor-LSTM-ARIMA model for predicting carbon emissions, utilizing tensor decomposition data analysis methods. The results indicate that this combined model effectively captures the complex relationships within heterogeneous data, outperforming baseline models in prediction accuracy. Furthermore, the study demonstrates that unstructured social media data can enhance structured time series data, providing a new perspective for comprehensively understanding the variables influencing carbon emission predictions.