A decision-centric framework for integrated energy and thermal management in sustainable systems: A simulation-based methodological study

Renewable-dominated energy systems increasingly depend on intelligent management that converts forecasts into operational decisions, yet the intermittency of renewable generation and the growing weight of thermally driven demand for heating and cooling continue to challenge stability, efficiency, and operational decision-making. Existing approaches remain fragmented: forecasting studies stop at prediction accuracy, thermal models are developed separately from renewable supply, and decision mechanisms are only loosely coupled to learned models. To close this gap, this paper presents the Agentic Energy–Thermal Decision Framework (AET-D 2 ), a unified, lightweight architecture that combines machine learning–based renewable energy forecasting with physics-informed thermal load estimation and introduces the Energy–Thermal Stress Index (ETSI) to quantify the balance between predicted energy availability and thermal demand, from which a transparent utility-based policy layer assigns control recommendations such as energy storage, normal operation, and HVAC-aware load adaptation. The study is a simulation-based methodological evaluation: in a reproducible setting driven by public weather data and a parameterized thermal proxy, ensemble models reach R 2 = 0.9075 on testing and R 2 = 0.9421 under cross-validation. Under the nominal policy coefficients, HVAC-aware load adaptation is assigned to 94.2% of the evaluated samples. Across the utility-based sensitivity configurations, thermal-side adaptation remains the dominant recommendation, although the split between HVAC moderation and load reduction is sensitive to the comfort-gap weights. These results demonstrate the internal behavior and reproducibility of the framework under the evaluated assumptions; validation on operational data is still required before drawing conclusions about control effectiveness or practical deployment.