A physics-informed machine learning framework for climate-aware digital twins in decentralised energy systems

The increasing integration of renewable energy resources in decentralised energy systems (DESs), particularly in weak-grid or isolated environments, has intensified operational uncertainty due to weather-driven variability. This variability often leads to renewable curtailment, insufficient system flexibility, and reliability concerns during extreme climatic events. Consequently, there is a growing need for forecasting and control approaches that maintain accuracy under changing climatic conditions while remaining reliable when available data are limited. Conventional physics-based models often struggle to capture the nonlinear and time-varying dynamics of DESs, whereas purely data-driven methods tend to lose predictive capability when exposed to unfamiliar climatic regimes. This study aims to improve multi-energy forecasting and operational control in DESs by developing a hybrid modelling framework that integrates Physics-Informed Machine Learning (PIML) with Climate-Aware Digital Twins (CADTs). The key novelty lies in enabling the digital twin (DT) architecture to dynamically adapt its internal state estimation and forecasting behaviour using climate-dependent inputs, while simultaneously enforcing thermodynamic and system-level physical constraints. The proposed CADT-PIML framework employs XGBoost models for photovoltaic and wind power forecasting, multilayer perceptrons to capture energy storage dynamics, and long short-term memory networks to predict electricity price fluctuations. Physical consistency is ensured through embedded thermodynamic constraints that guide the learning process and maintain interpretability. The proposed approach is validated through a rural microgrid case study to assess forecasting accuracy, operational reliability, and renewable energy utilisation. Results demonstrate that the CADT-PIML framework improves forecasting accuracy by 27.2%, reduces unmet load by 38.4%, and decreases renewable curtailment by 21.9% compared with conventional DT approaches. These findings indicate that climate-aware, physics-guided DTs can provide a robust and adaptive solution for intelligent decentralised energy management under increasingly variable climatic conditions.