Research on coal price prediction integrating XGBoost and graph convolutional network

SHAO Feng; FENG Yu; LU Yiguang; GENG Guoqiang; SHAO Hu

Abstract：Coal,as a key pillar of the global energy structure,has widespread effects on the economy,energy markets,environmental policies,and industrial production costs due to its price fluctuations.Accurately predicting coal prices is crucial for maintaining energy market stability,effectively controlling costs,and managing risks.This paper introduces a deep learning model—Extreme Gradient Boosting-Graph Convolutional Network(XGBGCN)—to address the problem of coal price prediction.The model integrates the Extreme Gradient Boosting(XGBoost)algorithm to analyze critical features affecting coal prices and the Graph Convolutional Network(GCN)model to predict prices based on these features.The XGBoost model can effectively extract key features from a large number of factors related to coal prices,thereby reducing model complexity and improving prediction accuracy.Specifically,the XGBGCN model first uses the XGBoost algorithm to identify features highly correlated with coal prices,such as electricity consumption,coal prices in other regions,etc.It then constructs an adjacency matrix of the coal price correlation graph using the selected features,combined with the feature matrix,as input to the GCN model for coal price prediction.Additionally,this paper predicts coal prices on an actual coal price and its influencing factors dataset,and the results show that the proposed XGBGCN model can accurately predict coal prices and outperform several existing models.