The Iberian Blackout

Module 5: Complex Networks

Reading

Compulsory: Motter and Lai (2002).
Recommended: Schäfer et al. (2018).

This lab studies cascading failures with a network model. It starts from a real event so you can see, from the start, how much a simple model leaves out.

What happened on 28 April 2025

The facts below come from the two reports of the Expert Panel set up by ENTSO-E, the European association of transmission system operators: the factual report of 3 October 2025 (ENTSO-E ICS Investigation Expert Panel 2025) and the final report of 20 March 2026 (ENTSO-E ICS Investigation Expert Panel 2026). All times are CEST.

At 12:33 the power systems of continental Spain and continental Portugal lost their entire supply. A small area of France near the Spanish border had disruptions for a limited time. The rest of the Continental Europe system had no significant disturbance (ENTSO-E ICS Investigation Expert Panel 2026, sec. 1.1).
The incident was classified at scale 3, the highest severity level of the ENTSO-E Incident Classification Scale (ENTSO-E ICS Investigation Expert Panel 2026, sec. 1.1).
In the half hour before the blackout, the grid showed two oscillation episodes: a local oscillation at about 0.63 Hz (12:03 to 12:08) and an inter-area oscillation at about 0.2 Hz (12:19 to 12:22) (ENTSO-E ICS Investigation Expert Panel 2026, sec. 1.2).
The final report concludes that the blackout resulted from multiple interacting factors. The sequence combined voltage fluctuations and oscillations, widespread generation disconnections in Spain (particularly of inverter-based resources), a cascade of overvoltage disconnections, and the loss of synchronism of the Iberian system with Continental Europe (ENTSO-E ICS Investigation Expert Panel 2026, sec. 1.9).
The Iberian systems started to lose synchronism with the rest of Europe at 12:33:19. Automatic load shedding and system defence plans activated between 12:33:19 and 12:33:22 but could not prevent the collapse (ENTSO-E ICS Investigation Expert Panel 2026, sec. 1.3).
The analysis indicates that the blackout originated in Spain, not in Portugal (ENTSO-E ICS Investigation Expert Panel 2026, sec. 1.4).
Restoration was completed in Portugal by 00:22 on 29 April. In Spain the transmission system was fully restored by 04:00 on 29 April (ENTSO-E ICS Investigation Expert Panel 2026, sec. 1.1).

The reports contain a root cause tree and a long list of recommendations. Read the management summary of the final report if you want the full picture. This page does not try to summarize the causes beyond the points above.

What a network model can and cannot say

The word “cascade” appears in the official conclusion, and that is the only link to this lab. The cascade on 28 April involved voltage control, reactive power, oscillations, protection settings and the behaviour of generating units (ENTSO-E ICS Investigation Expert Panel 2026). None of these quantities exists in the model you will build.

The Motter-Lai model (Motter and Lai 2002) is an intentionally simplified cascade model:

the grid is a graph with no electrical parameters;
the “load” of a node is the number of shortest paths through it;
a node fails when its load exceeds a fixed capacity;
there is no time, no voltage, no frequency and no control action.

It is not an explanation of the Iberian blackout. It is a minimal model that isolates one mechanism: when a node fails, its load moves to other nodes, and some of those may fail in turn. Use it to ask how network topology and spare capacity change the size of a cascade. Models that include the power flow dynamics exist, for example Schäfer et al. (2018); the Swing Equation page gives a first look.

Prediction

Exercise

Before you open the next page, write down your answers:

A network has a few hubs that carry most of the shortest paths. Do you expect it to be more fragile when a random node fails or when its most loaded node fails?
Every node gets a capacity 10% above its initial load. Do you expect a small or a large cascade? What about 100%?

You will test both predictions in Cascade Simulation.

Solution

When its most loaded node fails. A random node is almost always a node with low load, so little load moves. The most loaded hub sends a large share of all shortest paths to other nodes at once.
With a 10% margin a large cascade is likely, at least when the trigger is a highly loaded node. With a 100% margin most triggers cause only small cascades, but on a network with strong hubs the most loaded node can still do damage. The figure in Cascade Simulation shows both effects.

References

ENTSO-E ICS Investigation Expert Panel. 2025. “Grid Incident in Spain and Portugal on 28 April 2025: ICS Investigation Expert Panel Factual Report.” Brussels: ENTSO-E. https://www.entsoe.eu/publications/blackout/28-april-2025-iberian-blackout/.

———. 2026. “Grid Incident in Spain and Portugal on 28 April 2025: ICS Investigation Expert Panel Final Report.” Brussels: ENTSO-E. https://www.entsoe.eu/publications/blackout/28-april-2025-iberian-blackout/.

Motter, Adilson E., and Ying-Cheng Lai. 2002. “Cascade-Based Attacks on Complex Networks.” Physical Review E 66 (6): 065102. https://doi.org/10.1103/PhysRevE.66.065102.

Schäfer, Benjamin, Dirk Witthaut, Marc Timme, and Vito Latora. 2018. “Dynamically Induced Cascading Failures in Power Grids.” Nature Communications 9: 1975. https://doi.org/10.1038/s41467-018-04287-5.

--- title: "The Iberian Blackout" subtitle: "Module 5: Complex Networks" format: html --- ::: {.callout-note} ## Reading - **Compulsory:** @motter2002cascade. - **Recommended:** @schafer2018dynamically. ::: This lab studies cascading failures with a network model. It starts from a real event so you can see, from the start, how much a simple model leaves out. ## What happened on 28 April 2025 The facts below come from the two reports of the Expert Panel set up by ENTSO-E, the European association of transmission system operators: the factual report of 3 October 2025 [@entsoe2025factual] and the final report of 20 March 2026 [@entsoe2026final]. All times are CEST. - At 12:33 the power systems of continental Spain and continental Portugal lost their entire supply. A small area of France near the Spanish border had disruptions for a limited time. The rest of the Continental Europe system had no significant disturbance [@entsoe2026final, Section 1.1]. - The incident was classified at scale 3, the highest severity level of the ENTSO-E Incident Classification Scale [@entsoe2026final, Section 1.1]. - In the half hour before the blackout, the grid showed two oscillation episodes: a local oscillation at about 0.63 Hz (12:03 to 12:08) and an inter-area oscillation at about 0.2 Hz (12:19 to 12:22) [@entsoe2026final, Section 1.2]. - The final report concludes that the blackout resulted from **multiple interacting factors**. The sequence combined voltage fluctuations and oscillations, widespread generation disconnections in Spain (particularly of inverter-based resources), a cascade of overvoltage disconnections, and the loss of synchronism of the Iberian system with Continental Europe [@entsoe2026final, Section 1.9]. - The Iberian systems started to lose synchronism with the rest of Europe at 12:33:19. Automatic load shedding and system defence plans activated between 12:33:19 and 12:33:22 but could not prevent the collapse [@entsoe2026final, Section 1.3]. - The analysis indicates that the blackout originated in Spain, not in Portugal [@entsoe2026final, Section 1.4]. - Restoration was completed in Portugal by 00:22 on 29 April. In Spain the transmission system was fully restored by 04:00 on 29 April [@entsoe2026final, Section 1.1]. The reports contain a root cause tree and a long list of recommendations. Read the management summary of the final report if you want the full picture. This page does not try to summarize the causes beyond the points above. ## What a network model can and cannot say The word "cascade" appears in the official conclusion, and that is the only link to this lab. The cascade on 28 April involved voltage control, reactive power, oscillations, protection settings and the behaviour of generating units [@entsoe2026final]. None of these quantities exists in the model you will build. The Motter-Lai model [@motter2002cascade] is an **intentionally simplified** cascade model: - the grid is a graph with no electrical parameters; - the "load" of a node is the number of shortest paths through it; - a node fails when its load exceeds a fixed capacity; - there is no time, no voltage, no frequency and no control action. It is **not an explanation of the Iberian blackout**. It is a minimal model that isolates one mechanism: when a node fails, its load moves to other nodes, and some of those may fail in turn. Use it to ask how network topology and spare capacity change the size of a cascade. Models that include the power flow dynamics exist, for example @schafer2018dynamically; the [Swing Equation](swing-equation.qmd) page gives a first look. ## Prediction ::: {.callout-note} ## Exercise Before you open the next page, write down your answers: 1. A network has a few hubs that carry most of the shortest paths. Do you expect it to be more fragile when a random node fails or when its most loaded node fails? 2. Every node gets a capacity 10% above its initial load. Do you expect a small or a large cascade? What about 100%? You will test both predictions in [Cascade Simulation](cascade-simulation.qmd). ::: ::: {.callout-tip collapse="true"} ## Solution 1. When its most loaded node fails. A random node is almost always a node with low load, so little load moves. The most loaded hub sends a large share of all shortest paths to other nodes at once. 2. With a 10% margin a large cascade is likely, at least when the trigger is a highly loaded node. With a 100% margin most triggers cause only small cascades, but on a network with strong hubs the most loaded node can still do damage. The figure in [Cascade Simulation](cascade-simulation.qmd) shows both effects. :::