Any questions?

Your partner for machine learning in the energy industry:

Georg Fischer

Managing Advisor


Artificial Intelligence for Operational Excellence | forecast of highly volatile transport volumes

At the centre of the project is the Trans-Austria Gas Pipeline (TAG). It transports natural gas between the Austrian-Slovakian border in Baumgarten and the Austrian-Italian border in Arnoldstein/Carinthia. The gas flows that occur in the process are becoming increasingly volatile and forecasts are becoming more and more difficult. Together with WECOM, TAG GmbH laid the basis for its corporate AI strategy and developed a successful prototype for time series forecasting.

Client Profile

  • Trans Austria Gasleitung GmbH
  • gas transmission system operator
  • > 150 employees
  • 1.150 km pipeline network
  • up to 30 bcm/a gas transport volume

TAG GmbH is Austria's largest gas transmission system operator. Its core task is the operation and maintenance of gas pipelines and compressor stations to secure the energy supply in Austria and Italy.

The aim of the TRAIAN (TRAnsport to Italy ANticipation) project was to explore, together with WECOM, the potential use of Artificial Intelligence and Machine Learning to increase efficiency and reduce emissions in the following setting:

  • Identification of use cases for Artificial Intelligence and Machine Learning methods, on the one hand for emission reduction in compressor operation and on the other hand for optimisation of energy purchase, capacity marketing and maintenance.
  • Estimation of the economic value added for these use cases.
  • Evaluation of the realisable improvements in the forecasting of transport demand ("energy flow forecasting") by using modern machine learning methods for time series forecasting.



Traditional methods are increasingly struggling to predict the transport demands necessary for the energy-efficient and cost-effective operation of the transmission system. This is due to the fact that customer demands have to be taken into account at ever shorter notice in response to prevalent temperature fluctuations, market prices and storage usage conditions. At the same time, the demands on the operation of the transmission system are growing in terms of energy efficiency, marketing of transport capacities and efficient maintenance, as well as emission reduction. The goal of the Italian Snam S.p.A., the majority stake-holder of TAG GmbH, to operate CO₂-neutral by 2040 is particularly ambitious.

Specifically, the following challenges emerged during the project:

  • Complex modelling due to high volatility and lack of trends and seasonality in time series data.
  • Data standardisation, as the required data was distributed across different IT systems in a non-uniform format.
  • Low number of reference projects ("best practices") due to the novelty of machine learning methods in the energy industry.



First, the possible areas of application of AI/ML at a TSO (Transmission System Operator) were identified and subsequently the current and expected future potential of these applications was evaluated, taking into account changing framework conditions.
On this basis, a technical prototype for the particular use case of "energy flow forecasting" was developed as a company flagship project for AI/ML. Additionally, an economic feasibility analysis was carried out.


As a result, the customer's management is now in a position to formulate an AI strategy and decide on the implementation of machine learning methods for operational optimisation.
Especially for the use case "Energy Flow Forecasting", a significant improvement of the forecast quality compared to the reference method in relevant KPIs could be achieved. This confirms the economic potential, under consideration of the compressor operation as well as the corresponding electricity and gas prices.


  • 2.3 mio

    datapoints analyzed and processed

  • 5

    different AI/ML algorithms along with 6.500 models

  • 4 k

    lines of code created and tested in python

  • 150

    experiments evaluated under different metrics

Back to the news overview

Further Case Studies