LNBi-GRU model for coal price prediction and pattern recognition analysis

Abstract

Accurately predicting coal prices and identifying related factors are of great significance for energy market. However, limited attention has been devoted to the precision of predicting coal price trends and the comprehensive analysis of influencing factors. In this paper, we propose LNBi-GRU (Layer Normalization and Bidirectional GRU), which integrates Layer Normalization (LN) and Bidirectional network (Bi) to form the LNBi layer, thereby advancing coal price forecasting. Meanwhile, this study also achieves coal price pattern recognition through a combination of prediction, evaluation, and explanation. The results show that LNBi-GRU outperforms the selected baseline models in terms of both prediction accuracy and stability, and can predict mutation points more accurately. Ablation experiments prove the effectiveness of the added modules including LN and Bi. Moreover, market price emerges as a critical factor affecting coal prices. From a cyclical perspective, the dominant factor in the up cycle shifts from cost push to demand pull, while the dominant factor in the down cycle shifts from demand pull to cost push.