Intelligent coal price prediction method and application driven by multimodal data
WANG Ying
ZHANG Ruohan
ZHU Hongda
YANG Yang
ZU Zishuai
Abstract：Coal is a cornerstone of China's energy structure,and fluctuations in its price not only af-fect the upstream and downstream of the industrial chain but also bear on achieving the"dual-carbon"goals and safeguarding macroeconomic stability.To improve forecasting accuracy,this study develops an Informer-based multimodal cross-attention forecasting model,using the Bohai-Rim 5 500 K thermal coal price as the prediction target.We build a multimodal dataset that inte-grates key drivers including raw coal output,coal imports and tariffs,geopolitical uncertainty,coal enterprise inventories,total thermal power generation,and the consumer price index.The numeri-cal modality is processed via linear mapping;the textual modality employs a RoBERTa encoder to extract semantic representations;and a cross-attention mechanism is introduced for cross-modal fu-sion between numerical sequences and textual information.Empirical results show that Informer performs better than multiple benchmark models on long-horizon forecasting;relative to simple con-catenation,introducing cross-modal cross-attention significantly improves predictive accuracy;the model also passes robustness tests under noisy text conditions.The proposed Informer-based multi-modal cross-attention model provides methodological support for scientific coal price forecasting and offers theoretical and practical value for energy-market research and related policy formulation.
Keywords：multimodal datacoal priceInformercross-attention
Publication Date：2026-03-31
Online Publishing Date：2026-04-08(First online date of this platform, not the publication date of the document)
Pages：13( 546-558 )