A Study of the Effect of Hydrogen Doping on Real-Time Natural Gas Load Forecasting Models

With the advancement of global energy transitions and carbon reduction targets, hydrogen energy is gaining increasing attention as a clean energy source, particularly in the natural gas sector, where hydrogen blending is considered a key strategy for achieving carbon reduction goals. However, the incorporation of hydrogen alters the physical properties of natural gas, leading to an increased gas load demand from consumers. Consequently, existing deep learning models for short-term load forecasting face greater challenges. This paper examines the impact of hydrogen blending on the performance of deep learning load forecasting models, including BiGRU, BiLSTM, and RNN, with a focus on forecasting accuracy before and after hydrogen blending. The results indicate that hydrogen blending improves prediction accuracy in low and medium load ranges, but increases prediction errors in high-load zones and extreme load conditions, thereby reducing the model's adaptability. Among the models examined, the BiGRU model demonstrates the best performance in predicting hydrogenblended natural gas load, with potential for further optimization.