Carbon emission allowances, as the primary trading product in carbon markets, possess unique attributes. Their transaction prices serve as benchmark prices for carbon finance activities and aid in monitoring volatility risks within carbon trading markets. However, national policies and events in related markets can exert complex influences on carbon emission allowance trading prices. To capture these difficult-to-quantify unstructured textual insights and enhance the accuracy of carbon emission allowance price forecasting, this paper proposes a multi-source data feature fusion prediction model incorporating cross-attention mechanisms through the BERT-BiGRU-CNN-LSTM-CrossAttn multi-path feature channel. This model comprehensively integrates structured trading data and unstructured carbon text information influencing prices. Utilising historical carbon emission rights transaction data, relevant market supply-demand indicators, macroeconomic indicators, environmental climate metrics, and related news headlines as source data, it employs the BERT (Bi-directional Encoder Representations from Transformers) model for sentiment feature analysis. A combination of recurrent neural networks and convolutional neural networks—BiGRU (Bidirectional Gated Recurrent Unit, BiGRU)-CNN (Convolutional Neural Network, CNN) modules to extract numerical and textual features respectively. An LSTM (Long Short-Term Memory, LSTM) model incorporating a dynamic feature weighting mechanism is employed to forecast carbon emission trading prices. Empirical research was conducted using data from 2,506 trading days at the Hubei Provincial Carbon Emission Trading Centre between 2 April 2014 and 14 February 2025. Results indicate that incorporating textual features, sentiment features, and cross-attention mechanisms simultaneously significantly enhances prediction accuracy across various feature mechanism schemes. Furthermore, the predictive capability demonstrates clear advantages across different pilot cities and time windows.

As first or corresponding author, Professor Zhang has published more than 200 academic papers, with over 80 appearing in prestigious international SCI, SSCI, and A&HCI journals. He has received numerous academic honors, including being selected for the Shaanxi Sanqin Talents Special Support Program (Leading Talent in Philosophy and Social Sciences), the Ministry of Transport Young Science and Technology Talents Development Program, and the Shaanxi Provincial Outstanding Young Talents Support Program for Universities. Beyond his academic roles, he serves as a Legislative Consultant to the Standing Committee of the 14th Shaanxi Provincial People's Congress and Chief Expert of the Decision-making Advisory Expert Team of the Shaanxi Association for Science and Technology.