Machine learning of power grid frequency dynamics and control - prediction, explanation and stochastic modelling

Aug 5, 2026

Upload by Patrick

Reliable electricity supply depends on continuous balancing of power generation and demand. Grid frequency directly reflects this balance, and load-frequency control counteracts deviations to maintain stability. The shift toward renewables—characterized by low intrinsic inertia, weather dependence, and stronger coupling across economic sectors—creates faster, non-autonomous, and stochastic frequency dynamics. This thesis develops machine-learning tools to predict, explain, and model frequency behavior by integrating techno-economic drivers and stochastic dynamical modeling with validation and interpretable methods.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages132

Reading Time5 h 33 min

AccessFree to download

Abstract

Motivation and challenge

Proposed machine-learning approach

Prediction and explainability

Stochastic dynamical modelling

Machine learning of power grid frequency dynamics and control: prediction, explanation and stochastic modelling Inaugural-Dissertation zur Erlangung des Doktorgrades der Mathematisch-Naturwissenschaftlichen Fakultt der Universitt zu Klnvorgelegt von Johannes Kruse aus Hamburg Kln, 2023 ii Berichterstatter: Prof. Dr. Dirk Witthaut Prof. Dr. Simon Trebst Prof. Dr. Pere Colet Vorsitzende der Prfungskomission: Prof. Dr. Johannes Berg Tag der mndlichen Prfung: 27 . Januar 2023 Abstract A reliable supply of electric power is not a matter of course. Power grids enable the transport of power from generators to consumers, but their stable operation constantly requires corrective measures and a careful supervision. In particular, power generation and demand have to be balanced at all times. A large power imbalance threatens the reliability of the power supply and can, in extreme cases, lead to a large-scale blackout. Therefore, the power imbalance is constantly corrected through distinct control schemes. The power grid frequency measures the balance of power generation and demand. To guarantee frequency stability, and thereby a balance of generation and demand, loadfrequency control constantly counteracts large frequency deviations. However, the transition of the energy system to renewable energy sources challenges frequency stability and control. Wind and solar power do not provide intrinsic inertia, which leads to increasingly fast frequency dynamics. Different economic sectors become strongly coupled to the power system, as, for example, the adoption of electric vehicles will interconnect the transport sector and the power system. Finally, wind and solar power are weather-dependent, which increases the variability of power generation. All in all, this gives rise to diverse, interdependent and stochastic impact factors, that drive the balance of power demand and generation, and thus the grid frequency. How can we predict, explain and model frequency dynamics given its strong non-autonomous and stochastic character? In this thesis, I use machine learning to disentangle the effects of external drivers on grid frequency dynamics and control. First, I propose a prediction model that only uses historic frequency data, but fails in representing external impacts. Therefore, I include time series of techno-economic drivers and model their impact on grid frequency data using explainable machine learning methods. These methods reveal the dependencies between external drivers and frequency deviations, such as the important impact of forecast errors in the Scandinavian grid or the varying effects of different generation types. Finally, I integrate these drivers into a stochastic dynamical model of the grid frequency, which both represents short-term dynamics and long-term trends due to techno-economic impacts. My work complements traditional simulation-based approaches through validation and modelling inspiration. It offers flexible modelling and prediction tools for power system dynamics, which are profitable for systems with diverse impact factors but noisy and insufficient data. Zusammenfassung Eine zuverlssige Stromversorgung ist keine Selbstverstndlichkeit. Stromnetze ermglichen die ¨Ubertragung elektrischer Energie von den Erzeugern zu den Verbrauchern, aber ihr stabiler Betrieb erfordert stndige Korrekturmaßnahmen und eine sorgfltige ¨Uberwachung. Insbesondere mssen Stromerzeugung und-nachfrage jederzeit im Gleichgewicht sein. Ein großes Leistungsungleichgewicht gefhrdet die Zuverlssigkeit der Stromversorgung und kann im Extremfall zu einem großflchigen Stromausfall fhren. Daher wird das Leistungsungleichgewicht stndig durch verschiedene Regelungssysteme korrigiert. Die Netzfrequenz ist ein Maß fr das Gleichgewicht von Stromerzeugung und Stromnachfrage. Um die Frequenzstabilitt und damit ein Gleichgewicht zwischen Erzeugung und Nachfrage zu gewhrleisten, wirkt die Regelleistung stndig großen Frequenz

FAQs

Why is power grid frequency stability important?

Frequency stability ensures that generation and demand remain balanced. Large imbalances can threaten reliability and even lead to large-scale blackouts.

What makes modern frequency dynamics more difficult to control?

Renewables such as wind and solar provide little intrinsic inertia, and their weather dependence increases variability. Additional coupling between sectors further creates non-autonomous and stochastic influences.

How does the thesis use machine learning to address frequency dynamics?

It builds a prediction model from historical frequency data, then incorporates techno-economic driver time series and uses explainable machine learning to identify dependencies and driver effects. Finally, it integrates drivers into a stochastic dynamical model to capture both short-term dynamics and long-term trends.

Aug 5, 2026

Upload by Patrick