Decoding the energy paradox: An interpretable XGBoost-SHAP framework for forecasting decoupled demand futures

• A novel hybrid XGBoost-Prophet framework is developed for 30-year ahead hourly forecasting of electricity and natural gas demand. • The model explicitly incorporates population decline scenarios alongside climate change, addressing a critical research gap. • SHAP analysis ensures interpretability, revealing key drivers and complex, non-linear feature interactions. • Even mild climate warming can offset the demand-reducing effect of population decline on electricity consumption. • A structural divergence between electricity and natural gas demand trajectories is uncovered under future scenarios. • The generated high-resolution load profiles are a vital tool for de-risking investments in renewables and optimizing multi-energy systems. Effective energy system planning in cold-climate, heating-dominated regions necessitates accurate, high-resolution (hourly), and long-term energy demand forecasts. However, existing research gaps include the systematic evaluation of advanced machine learning methods for multi-decade horizons, explicit modeling of population decline, and simultaneous forecasting of integrated electricity and heat (natural gas) demand. This study fills these gaps by developing a novel hybrid framework to forecast hourly electricity and natural gas demand over a 30-year horizon in a representative cold-climate region. The framework utilizes XGBoost for hourly predictions and Prophet for annual trend correction. Model transparency is ensured via SHAP analysis. Results demonstrate high model generalizability, with R² values of 0.91 and 0.87 for electricity and natural gas, respectively. A key finding is that even mild climate warming can offset the demand-reducing effect of population decline on electricity consumption due to increased cooling needs. Furthermore, a structural divergence emerged: the demand-suppressing effect of severe climate warming on natural gas can become as potent as rapid population decline in the long term. This highlights profound interactions between climatic, demographic, and economic drivers. The generated hourly load profiles provide a vital, interpretable tool for integrated energy system planning, de-risking investments in renewables, and optimizing the transition toward a sustainable energy future.