System-level heat demand forecasting and operational optimization in a district heating network with large-scale TES by AI approach

Accurate forecasting and optimization of heat demand are essential for enhancing the operational efficiency and flexibility of district heating networks, especially when integrating large-scale thermal energy storage (TES). This work presents a novel framework that combines exploratory data analysis, automated machine learning (AutoML), and explainable AI (XAI) to enable robust prediction and operational optimization of heat demand in a city-scale district heating system equipped with a 12,000 m3 hot-water accumulator. Using a dataset of 8760 hourly operational records, the framework leverages a Gradient Boosting Regressor (GBR) and SHAP values to achieve high forecasting accuracy (R2 = 0.995, RMSE = 0.488 MW), identify key operational drivers, and provide interpretable decision support. Optimization studies using Bayesian methods (Optuna) reveal strategies that maximize system heat output under real-world constraints. The results support improved energy management, facilitate demand-side management and the integration of renewable energy, and contribute to the development of sustainable, energy-efficient district heating systems aligned with net-zero targets. The proposed model demonstrates strong generalizability and practical applicability for predictive control and strategic planning in modern energy systems.