Building public market opinion indices for electricity consumption prediction of Chinese nonferrous metal industry

Abstract

Nonferrous metal is an essential natural resource for national economy and industry production. Understanding the supply, demand and price trend of nonferrous metal market is important to economic prosperity analysis and might contribute to electricity power consumption forecast. This research builds related market opinion indices to understand the supply, demand, and price changes in nonferrous metal market. We propose a self-learning graph neural network model-based sentiment index network method for value mining of sentiment-related features, addressing the issue of sentiment loss caused by discontinuities in domain-specific news. Based on the structured indicators and the integration of Internet sentiment indices, we establish two single-item models: Granger Causality-based ARDL Model (G-ARDL) and the Factor Analysis-based Transformer Model (F-Transformer). Finally, we construct a Bagging random sampling integration prediction framework to obtain the weight coefficients for single model predictions. We merge the prediction results in different single models with weighted fusion, resulting in smaller prediction errors compared to single models and equal-weight integration models.