A hybrid attention-enabled multivariate information fusion for carbon price forecasting

Carbon emissions trading is crucial for advancing a country’s shift toward green and low-carbon development, which serves as a key market mechanism that encourages both enterprises and nations to work toward the “dual carbon” goal. Accurately predicting carbon prices is highly important for devising effective market strategies and environmental planning. This study introduces a deep neural network framework, namely CPFNet, to achieve accurate carbon price predictions. CPFNet leverages a hybrid attention mechanism to effectively capture and integrate feature representation of multivariate information, thereby improving the accuracy of carbon trading price information. Extensive experiments demonstrate that CPFNet significantly outperforms other state-of-the-art models in predicting carbon prices across numerous cities and historical horizons. Moreover, the attention mechanism quantifies the influence of multivariate information on carbon prices, providing valuable insights for carbon-related policy formulation, economic decision-making, risk management, etc.