Analysing grid stability under increasing penetration of renewables using machine learning techniques

Modern electricity grids are introducing more converter-connected renewable energy generators, which help combat climate change. These generators have zero carbon emissions of Scope 1 (instant emissions from burning fossil fuels) and have only Scope 3 carbon emissions (life-cycle emissions due to manufacturing, transportation, deployment and decommissioning). They can connect to the grid almost instantly, which is a great benefit for customers. However, most converter-connected generators lack the inherent grid stabilising functions of conventional generators, which have high inertia in their hardware (steam turbines powered by fossil fuels or nuclear power). As a result of this lower inertia, regular grid disturbances can lead to fast-changing grid frequency variations, potentially cascading to blackouts if uncontrolled. The goal of the modern energy systems is to combine the benefits of heterogeneous energy network with large penetration of renewable generators and highest possible stability of the grid at the level of the standard frequency 50Hz.

 

We analyse frequency data of the UK grid and demonstrate how the machine learning tools help automate anomaly detection. We apply Bayesian change point analysis and tipping point analysis (early warning signals) to identify and align anomalies, which require additional processing for precise timing of the anomaly events [1]. We also apply clustering to more than 300 datasets of the UK frequency anomaly events and identify several types of such anomalies.

 

Our results provide the initial grounds for automated preventive management of the grid stability under the increasing number of renewable generators in the national grid. The methodology is generic and can be applied to other types of datasets across Europe.

References

[1] Livina et al, Geoscientific Instrumentation, Methods and Data Systems 14, 541-564, 2025