The paper ‘Continuous estimation of power system inertia using convolutional neural networks', written by a group of researchers in Electrical Engineering from the Department of Electronics, Information and Bioengineering at the Politecnico di Milano in collaboration with Terna Rete Italia, was published in Nature Communications and selected among the journal’s Editors’ Highlights in the ‘AI and machine learning’ research area.

The study, conducted by Daniele Linaro, Federico Bizzarri, Davide del Giudice, Samuele Grillo e Angelo Brambilla, starts from the common assumption that inertia is a measure of a power system’s capability to counteract frequency disturbances: in conventional power networks, inertia is approximately constant over time, which contributes to network stability.

However, as the share of renewable energy sources increases, the inertia associated to synchronous generators declines, which may pose a threat to the overall stability. Reliably estimating the inertia of power systems dominated by inverted-connected sources has therefore become of paramount importance.

The paper develops a framework for the continuous estimation of the inertia in an electric power system, exploiting state-of-the-art artificial intelligence techniques. An in-depth investigation is then performed based on power spectra analysis and input-output correlations to explain how the artificial neural network operates in this specific realm, thus shedding light on the input features necessary for proper neural-network training. The approach is validated on a heterogeneous power network comprising synchronous generators, static compensators and converter-interfaced generation. The results highlight how different devices are characterized by distinct spectral footprints - a feature that must be taken into account by transmission system operators when performing online network stability analyses.