Following our previous blog on the newly published paper Intelligence-Based Prediction of Coefficient of Performance for a Novel High-Temperature Industrial Heat Pump, which demonstrated how artificial intelligence can accurately forecast the performance of Stirling-cycle systems, we now take the story further.

In this second installment, we speak directly with the lead author Iman Golpour, a post-graduate engineer at the National University of Distance Education (UNED), to unravel the thinking behind the research published in Energy and Buildings (Elsevier), April 2026.

As industries look to deploy high-temperature heat pumps more widely, understanding which AI models perform best, and why, delivers clear industrial value. Timing operation to favourable process and energy price conditions cuts costs, eases grid pressure, reduces environmental impact and even supports planned maintenance.

In this follow-up, Iman Golpour explains the AI models used in the study, Artificial Neural Networks (ANNs) and Adaptive Neuro-Fuzzy Inference Systems (ANFIS). He explains why ANN proved to be more accurate, and why Stirling-cycle industrial heat pumps are better suited to AI models than more unpredictable systems.

Why did you conduct the study?

Industrial heat pumps are expensive, complex, and are expected to run reliably at very high temperatures. If we can predict performance accurately, operators can design better systems, run them smarter, and avoid wasting electricity, money, and patience.

Why is this beneficial for delivering high-temperature process heat?

Because the Stirling heat pump can follow industrial demand and operate efficiently at very high temperatures, it is especially suitable for this kind of “charge when smart, deliver when needed” strategy.

The predictive model tells operators exactly how efficient the system will be at different conditions, so decisions are based on data, not guesswork.

For industrial stakeholders, the findings are particularly relevant as high-temperature process heat demands stable operation, high efficiency and intelligent timing.

By applying predictive models, operators can optimise when and how the heat pump runs — aligning operation with the cheapest or greenest electricity, avoiding inefficient conditions, and delivering the same thermal output with lower power consumption. That means lower costs, lower emissions, and less stress on the grid.

In practice, this could mean:

The heat pump runs more when electricity is cheap or renewable.
It avoids inefficient operating points.
Operators know the performance parameters before switching it on.

This helps:

Reduce electricity/CO2 emissions.
Support grid stability.
Improve planning and maintenance.

Why is the Stirling-cycle heat pump especially suited for this?

Stirling heat pumps can deliver high-temperature industrial heat electrically and they work at higher temperatures than most conventional heat pumps. In addition:

They do not have phase-change complications like other systems.
They show smooth, predictable behaviour when operating steadily.
Their performance can be predicted and optimized using data.

But it’s not just about prediction. It’s about confidence, scalability, and smart operation, allowing the integration of flexible, low-carbon heat where fossil systems are still dominant.

That makes the Stirling cycle ideal for AI models, which like systems that behave consistently, instead of chaotically.

Comparatively, conventional heat pumps:

Can struggle at very high temperatures;
Lose efficiency quickly; and
Are hard to control precisely.

What were the major findings?

The AI could “guess” the heat pump efficiency almost exactly before it runs. Essentially:

Both AI models could predict performance well.
The ANN model was extremely accurate, almost perfectly matching real measurements.
ANN performed better than ANFIS for this system.

Give us an comparative scenario that we can understand.

Imagine teaching two students to predict how well a machine will work.

Student A (ANN) practices a lot and learns patterns really well.
Student B (ANFIS) uses rules but gets confused sometimes.

Both do okay, but Student A is almost always right.

That’s what happened here. The ANN learned the system behaviour so well that its predictions were nearly perfect.

Why was the prediction so accurate?

· The system was tested in real industrial conditions.

· The input data directly affects performance.

· The ANN can learn non-linear relationships better.

Basically, the AI wasn’t guessing. It understood how the machine behaves.

What further studies are needed?

Testing with larger datasets.
Including dynamic operating conditions.
Integrating AI with real-time control systems.
Comparing with other high-temperature heat pump technologies.

Basically: move from prediction → control → automation.

How far away is this from reality?

Not far away, up to five years, maybe a bit longer. The prediction models are nearly proven, with the estimation that real-time integration could be in a few years.

The research was conducted as part of the European SUSHEAT project, co-ordinated by the Energy Engineering Department at Spain’s National University of Distance Education (UNED). Enerin is a project partner.

Stirling HTHP ideal for AI-led pre-start performance prediction for savings