Til hovedinnhold
Norsk English

Hybrid analytics

Combining machine learning with analytical models may give the best from both worlds.


Machine learning models are flexible and powerful, but they often work as black boxes with high data requirements, and they do not provide insight about the underlying physical phenomena. On the other hand, real world data mainly come from sensors or manual measurements that can suffer from low quality or quantity, and the underlying system is important for the interpretation of the measurements. 

Hybrid approach combining physical models with machine learning have several advantages: 

1) It preserves the physical information

2) it improves the flexibility and precision of the model

3) the physical information may help compensate for the inferior data quality

We focus on the use of machine learning to gain knowledge about physical systems, known as model identification. And we focus on using physical models combined with machine learning, e.g. in physics-informed neural networks but also other methods.

This type of individually tailored hybrid approach requires a combination of strong domain knowledge and machine learning expertise, but results in robust semi-interpretable models and increased physical understanding. 

One focus area within hybrid analytics is ML-in-the-loop, which focuses on applying machine learning methods in a control loop. In industry, most processes and systems are controlled to achieve stable and high production or to reach some goal, either through automated control systems or human operators in the loop. For complex systems, developing optimal, robust and efficient controllers using traditional control theory is highly challenging. On the other hand, data collected from an active control loop is less varied and do not fully represent the full, underlying dynamics. A hybrid approach combining existing domain knowledge and control theory with information in data may result in better estimation and control.

Hybrid analytics is an active field of research and SINTEF is participating at forefront within practical applications. In addition, SINTEF has inhouse competence within many domains allowing for close collaboration between domain experts and machine learning experts.

We have tailored solutions for process industry control systems, energy market predictions, drilling risers, predictive maintenance of marine propulsion systems, amongst others.

Utforsk fagområdene