A Physics-Informed Machine Learning Framework for Climate-Aware Digital Twins

This paper introduces an innovative approach to improving the reliability and efficiency of decentralised energy systems, particularly those heavily dependent on renewable energy.

As communities increasingly rely on locally generated energy from sources such as solar panels, wind turbines, and battery storage, managing these systems becomes more complex and uncertain. Renewable generation is highly sensitive to weather and longer-term climate variability, making accurate forecasting and system control especially challenging—particularly in rural or weak-grid environments where data may be limited. Traditional physics-based models often struggle to capture the nonlinear dynamics of such systems, while purely data-driven approaches can fail when conditions fall outside their training experience.

This research addresses these challenges through the development of a novel CADT-PIML framework, which combines Physics-Informed Machine Learning (PIML) with Climate-Aware Digital Twins (CADTs). By embedding fundamental physical laws—such as thermodynamic and operational constraints—directly into machine learning models, the framework is able to retain physical realism while benefiting from the adaptability and predictive power of artificial intelligence. At the same time, the digital twin component continuously updates its behaviour based on evolving climate conditions, allowing the system to respond not only to immediate weather fluctuations, but also to longer-term climate patterns such as prolonged heatwaves or wind droughts.

The framework was tested on a rural microgrid in Norway, providing a real-world demonstration of its capabilities. The results are striking: the system achieved a 27.2% improvement in energy forecasting accuracy, a 38.4% reduction in unmet energy demand, and a 21.9% decrease in renewable energy curtailment. These improvements translate into more reliable energy supply, better use of renewable resources, and reduced waste—key priorities for sustainable and resilient energy systems.

An important feature of the approach is its emphasis on transparency and trust. Using SHAP-based explainability tools, the framework provides clear, interpretable insights into how decisions are made. This enables system operators to understand and validate the model’s recommendations, addressing a critical barrier to the adoption of advanced AI-driven solutions in real-world infrastructure.

Overall, this work represents a significant step forward in the design of intelligent, climate-aware energy systems. By bridging the gap between physical modelling and machine learning, the framework offers a robust and adaptable solution for managing decentralised energy networks under increasingly uncertain climate conditions. It provides a strong foundation for future developments in digital twins and supports the transition toward smarter, more resilient local energy systems.

This research was supported by UK Research and Innovation (UKRI) under the SAT-Guard project (MR/Z50578X/1).

Authors: Muhammed Cavus, Jing Jiang, Adib Allahham, Hongjian Sun

Read the paper: https://doi.org/10.1016/j.apenergy.2026.128013

Code & Data: https://github.com/cavusmuhammed68/CADT-PIML

Read the paper: https://doi.org/10.1016/j.rser.2026.116947

Photo: appliedenergy.jpg