Integrating multi-modal data into transformer model for short-term gas consumption forecasting

Natural gas plays a critical role in our daily lives, driving social and economic development. Hence, accurate forecasting of natural gas consumption is essential in ensuring a reliable supply. The proliferation of natural gas monitoring devices, combined with advancements in deep learning technologies, has enabled significant progress in gas consumption forecasting. Traditional deep-learning algorithms often rely on raw data, lack effective feature extraction methods, and struggle to fully leverage the multimodal and multidimensional characteristics of gas consumption data. To address these issues, this study proposes the GPPL gas-consumption forecasting model, which employs principal component analysis (PCA) for feature extraction and dimensionality reduction to overcome the challenges posed by multidimensional gas consumption data. In addition, it incorporates a Transformer architecture to improve forecasting accuracy. Experimental results demonstrate that the comprehensive GPPL model achieves superior forecasting performance compared to baseline methods and current mainstream deep learning models.