A dual-layer modelling method for characterizing and optimizing the energy flexibility of building heating systems

Abstract

Utilizing passive thermal storage with electric heat pumps enables cost-effective building energy flexibility enhancement. Current reduced-order modeling approaches for control simulations introduce significant simulation-reality discrepancies, while high-fidelity physical models require impractical computational loads for real-time applications. To address these challenges, this study introduces a dual-layer modelling framework. An integrated detailed model comprising three interactive sub-models (multi-zone buildings, heating devices, heat pumps) is developed as a digital twin platform. Using the training data, an artificial neural network (ANN) model for predictive control is constructed and validated. Three strategies including proportional (PC), rule-based (RC), and model predictive control (MPC) are evaluated via four flexibility metrics: load shifting ratio, energy transfer efficiency, flexibility coefficient, and thermal discharge duration. Simulation results indicate that radiator heating systems achieve 2.9 times higher load shifting capacity versus fan coil systems. MPC with optimized simulated annealing (SA) algorithms reduces operational costs by > 50 % in mild conditions and 5 % during peak heating demand compared to RC strategies. Quantitative analysis demonstrates that climate variations and terminal configurations significantly influence flexibility potential.