Adaptive data analytics and modelling for flexible power systems

  • Mäki, Kari (PI)
  • Järventausta, Pertti (CoPI)
  • Niska, Harri (CoPI)
  • Koponen, Pekka (Participant)
  • Ikäheimo, Jussi (Participant)
  • Motta, Sergio (Participant)
  • Kiljander, Jussi (Participant)
  • Kuusela, Pirkko (Participant)
  • Koskela, Juha J. (Participant)
  • Hilden, Antti (Participant)
  • Pakonen, Pertti (Participant)
  • Mutanen, Antti (Participant)
  • Brester, Christina (Participant)

Project: Academy of Finland project

Project Details


The project studies and develops hybrid models that integrate different modelling methodologies such as physically based models and different data driven methods, including deep learning and other state of the art machine learning methods. Such hybrids combine the strengths and mitigate the weaknesses of the component approaches. The applications studied are related to power grids and include forecasting of loads, distributed generation, power flows and grid state, and analytics of big power quality data.

Continues the work started in SA Response 2005-2018 and SAISEI 2018 – 2019 that applied hybrid modelling to short term forecasting of the load control responses of active demand.

Leader: Prof. Kari Mäki, VTT
Leader of sub project: Prof. Pertti Järventausta, TUNI
Leader of sub project: Ph.D. Harri Niska, University, UEF
Funding: Academy of Finland, co-funded. Total cost: 1 210k€

Layman's description

In new or rare situations purely machine learning based methods tend to produce very inaccurate forecasts which may have expensive and undesirable consequences in the energy grids. Integrating machine learning and physical models the project can provide better models and forecasts. Accuracy, reliability and robustness of the forecasts improves. The project also develops hybrid analytics (combination of physically based models and state of the art machine learning) for efficiently extracting useful information from the huge amounts of data available from power quality monitoring, network automation, smart metering, external systems and public sources.

Key findings

1) Hybrid analytics can combine the strengths and mitigate the weaknesses of different modelling approaches. 2) Careful selection and application of performance criteria is important. 3) The uncertainties of the forecasts need to be modeled. 4) Power quality analytics need to be developed in many ways in order to efficiently use the related huge data that is being collected.
Short titleAnalytics
AcronymSA Analytics
Effective start/end date1/09/1931/08/23

Collaborative partners


  • Smart grids
  • energy networks
  • data analytics
  • machine learning
  • deep learning
  • load modelling
  • network simulation
  • power flow calculation
  • power quality