The integrated electricity-heat-hydrogen energy system is regarded as a vital enabler of future green energy development due to its clean and sustainable properties. Accurate and reliable forecasting of multi-energy loads, such as electricity, heat, and hydrogen, is critical for the optimal scheduling and efficient management of integrated energy systems. Considering the complex coupling relationships among multi-energy loads and uncertainties of external factors, generating sufficiently informative probability distributions of multi-energy loads becomes a critical challenge. First, to extract the primary trends and remove noise from the original loads, Complete Ensemble Empirical Mode Decomposition (CEEMD) is used to reconstruct electricity, heat, and hydrogen load signals. Then, to quantify the uncertainties and capture complex coupling relationships among multi-energy loads, a multi-task learning framework integrating Bayesian Neural Networks (BNN) and Long Short-Term Memory (LSTM) networks is employed. Finally, validation using real-world load datasets, compiled from field surveys and multi-source data integration, demonstrates that the proposed model outperforms traditional methods, improving accuracy metrics by at least 40% and reliability metrics by at least 6%. • Forecasting framework for Electricity-Heat-Hydrogen IES. • Hybrid CEEMD-MTL-BNN model handles noise and uncertainty. • Multi-task learning captures cross-energy coupling effects. • Probabilistic forecasts reduce MAPE by 41% on real data.
Wei et al. (Sun,) studied this question.