Deep learning for gas pipelines

Deep learning for gas pipelines



We were mandated to design, develop and industrialise an innovating gas quality assignment solution on a national level gas network based on hybrid Machine learning/Finite element modelling.


Following a study of the state of the art in question, we designed a modular, hybrid simulation solution that used a set of unit models and was capable of adapting to any topology in real time. The calculation core was developed using python and then optimised to the best of its possibilities using the Cython language. Once the real time simulation had been designed, we included many additional models in it (slow leak detection and targeting, data reconciliation for failing measurements, reconstitution of stocks in the pipeline, etc.). The solution was designed to be modular and is built into the client’s industrial AWS environment.


Significant gains
in solution accuracy and adaptability

Reduced risks
of simulation discrepancies in cases of measurement failures using measurement reconciliation

Line control
compared to the off the shelf solution initially used