AI-generated image using Gemini (Copyright: RTDS 2026)
Why smarter heat pumps matter
Industry is one of Europe’s largest energy consumers, with many industrial processes still relying on burning fossil fuels to generate process heat at high temperatures. High-temperature heat pumps offer a promising alternative by recovering waste heat and upgrading it for reuse, reducing both energy consumption and greenhouse gas emissions. However, to deliver these benefits, heat pumps must operate efficiently under changing real-world conditions. In addition, predicting their performance using traditional engineering calculations alone can be challenging, particularly for innovative technologies such as Stirling cycle high-temperature heat pumps. Unlike conventional heat pumps, Stirling cycle systems are specifically designed to deliver the high temperatures needed by many industrial processes, making them a promising solution for decarbonising industrial heating.
This is where artificial intelligence (AI) comes in. By learning from operating data, AI can predict how a heat pump will perform under different conditions, helping optimise system operation, support predictive maintenance and guide future system design. More accurate predictions can also help increase confidence in deploying industrial heat pumps at a larger scale.
In a new study, researchers from Spain’s National University of Distance Education (UNED) investigated how AI can predict the performance of the Enerin AS‘s novel Stirling cycle high-temperature heat pump (SC-HTHP). They compared two machine-learning models to determine which can estimate the system’s efficiency most accurately.
This work was conducted as part of the European SUSHEAT project, coordinated by the UNED. The study was carried out by researchers from UNED (Spain), Enerin AS (Norway), Wiz Research (Romania), and Lucian Blaga University of Sibiu (Romania).
Furthermore, it is based on experimental tests collected under real industrial operating conditions at the IVAR biogas plant in Stavanger, Norway. Published in Energy and Buildings in April 2026, it is titled Intelligence-based prediction of coefficient of performance for a novel high-temperature industrial heat pump: Comparative performance of ANN and ANFIS models.
AI-generated image using Canva AI (Copyright: RTDS 2026)
How AI Learns to Predict Heat Pump Performance (COP)
The Coefficient of Performance (COP) measures how efficiently a heat pump converts electricity into useful heat. The higher the COP, the less electricity is needed to produce the same amount of heat.
The researchers compared two types of machine-learning models to predict the COP that can identify patterns in data and learn the relationships between different variables.
The first was an Artificial Neural Network (ANN). Neural networks are a type of machine-learning model inspired by the way the human brain processes information. Rather than following fixed rules, they learn patterns by analysing large amounts of data and gradually improving their predictions during training.
The second was an Adaptive Neuro-Fuzzy Inference System (ANFIS). This approach combines neural networks with fuzzy logic, a method that mimics human reasoning by using flexible “if-then” rules instead of simple yes-or-no decisions.
To train both models, the researchers used four operating parameters measured during experiments: the average heat source temperature, the average heat sink temperature, the hot water inlet temperature, and the temperature ratio.
¨Both models demonstrated strong predictive capability based on the selected model configurations and input data. However, the ANN model achieved slightly higher overall accuracy,” said Iman Golpour of UNED, the corresponding author of the publication.¨
AI-generated image using Canva AI (Copyright: RTDS 2026)
Filling an important research gap
Despite being established for decades, AI has seen substantial growth in engineering applications during the past two decades. Consequently, AI-based performance prediction for heat pumps is already an active research area at the international level. For example, some researchers have successfully applied both different AI models to predict the COP of ground-source heat pumps with excellent accuracy.
However, the SUSHEAT authors from the UNED, ENERIN and WIZ Research, together with Lucian Blaga University of Sibiu, found that previous research on using AI models for predicting the COP of the Stirling cycle high-temperature heat pumps for industrial applications is limited. Moreover, direct comparisons of different machine-learning approaches for these systems are largely missing.
By combining experimental testing with AI-based modelling, the SUSHEAT project helps address this knowledge gap and advances the development of intelligent industrial heating technologies in Europe.
Which AI Model Performed Best?
Both AI models were able to estimate how efficiently the heat pump would operate under different conditions. However, the ANN consistently produced predictions that were closer to the actual performance measured during the experiments than the ANFIS model.
In this study, prediction accuracy refers to how closely the AI model’s estimated COP matched the COP measured during real-world testing. In other words, a more accurate model gives more reliable estimates of how efficiently the heat pump will convert electricity into useful heat under different operating conditions.
The researchers conclude that the ANN was better able to capture the complex relationships between the heat pump’s operating conditions and its performance. They also note that the ANFIS model may have been limited by the relatively small amount of training data available, and that its accuracy could improve with larger and more diverse datasets.
Overall, the results indicate that Artificial Neural Networks are particularly well suited for predicting the performance of Stirling cycle high-temperature heat pumps, making them a promising tool for improving the operation and optimisation of industrial heating systems.
Looking Forward
The researchers recommend that future work investigate additional operating factors, including different working fluids, and a wider range of pressure conditions, to improve model performance. They also point to the value of using more advanced machine learning methods, along with experimental testing across a wider range of conditions, to improve accuracy.
This study offers a practical example of how the technologies being developed within SUSHEAT can contribute to smarter industrial energy systems. As high-temperature heat pumps become increasingly important for industrial decarbonization, AI-based tools will help ensure they operate as efficiently and reliably as possible.