By now, most of us will have heard about the collapse of the Iberian Peninsula’s transmission system on 28 April 2025.

The relative mix of renewables to conventional generation, power system inertia, voltage effects, protection settings and atmospheric effects on power line electrical characteristics are among the aspects that are currently being examined. There are several excellent articles on Linkedin discussing the event, in particular the work of Professor Luis Badesa.

No definitive root cause that has been identified at this time, and analysis of the event is proceeding. However, it is clear that we are entering uncharted territory in grid supervision and operation, as the level of renewables penetration rises.

It is important to understand this event thoroughly and as soon as possible, as the issue can potentially occur elsewhere in the future. Many nations are well down the path of dominant renewable generation. Many are in a position where the grid is operated at 100% renewables for periods of time. Spain is one example, South Australia another. There will be several more internationally.

This is a clear and present risk and it needs to be managed immediately.

So what can be done in the short, medium and long term to protect against events such as this, or at least mitigate the risk of occurrence?

Here is one strategy, with timeframes:

    1    Revisit and revise where necessary, system security calibrations and contingency management (Short term)

    2    Revisit and revise where necessary, protection philosophies and protection calibration, in particular for under-frequency, over-frequency and rate of change of frequency protection (short to medium term), and

    3    Review and revise where necessary, the target mix of technologies, to ensure a functional level of intrinsic protection against, and recoverability from, system collapses (strategy: short to medium term, implementation: medium to long term)

System security is an established part of power system operations worldwide. There is an opportunity for improving mitigation of the risk of these events, merely by recalibrating the level of security that operations are to achieve.

System security is essentially the power system’s ability to withstand individual events such as the unexpected loss of a transmission line, substation or generator. Typically, security is assessed in a “what-if” manner, by examining single contingencies and assessing their impact on power, voltage and frequency.

Normally, single contingencies are considered in security assessment. However, multiple contingencies are often considered for more complex power systems. It follows that multiple contingencies create combinatorial increase, and so the security assessment requires significantly more computation. Combinatorial explosion occurs as the number of contingencies being assessed rises.

It may be that system security policy needs to change, to include multiple contingencies. This is something that can be done immediately, with most of the currently deployed power system application suites.

Artificial Intelligence (AI) and machine learning (ML) may assist in security assessment. For example, rather than randomly selecting groups of contingencies for assessment, AI and ML may be used to examine the combinations and prioritise them, from the most credible to the least credible, thereby focusing the security assessment on scenarios that would be more likely to occur.

One of the characteristics of the Iberian peninsula collapse was at least two periods of rapidly oscillating power. This can be seen on the system frequency graphs around the time of the event.

Power oscillations are normally seen at much lower frequencies. The high frequencies that were seen on 28 April quite likely triggered frequency protection at multiple locations, which in turn would have contributed to the instability of the system and its eventual collapse.

Frequency protection is the power system’s last line of defence against excessive mismatch between the instantaneous power being generated, and the corresponding instantaneous electrical load.

When load exceeds what is being generated, the frequency falls. Conversely, when too much generation is present, frequency rises.

Protection is placed at various locations on the power system to shed load when the frequency is too low, and shed generation when the frequency is too high. There is also a type of frequency protection which acts if the frequency falls or rises too quickly, to preemptively stop the change before it becomes excessive. (Rate of change protection).

Frequency protection is normally calibrated for relatively slow changes in comparison to what was seen on 28 April. Review of protection settings is something that can also begin in the short term. However, it would take time to implement, commission and validate any revisions given the distribution of this type of protection around the power system.

AI and ML can assist with the review and operation of frequency (and other types of) protection. For example, AI and ML can be used to identify patterns in the sequence of events that can yield insight into which areas of the power system need to be examined first. AI and ML “on the edge” may help by changing the behaviour of protection by examining other local parameters such as voltage, and momentarily delaying the activation of protection if the situation warrants it.

Finally, the review and revision of the final, target mix of technologies for the power system. Strategy and planning is a medium term activity, and implementation occurs over a much longer term. But it is important to start this now, for timely prevention of a poor outcome, if progress happens to be along a bad trajectory.

AI and ML can assist with this task as well. For example, heuristic optimisation can help to identify a better optimal technology mix in comparison to conventional optimisation.

Artificial intelligence and machine learning in power system supervision was first considered in the 1980’s. Back then, the problem was the attrition of personnel experienced in the operation and supervision of power systems. Nowadays, thanks to the advancement in the power, flexibility and scale of computing and communications technology, the potential applications of AI and ML extend well beyond this.

Artificial intelligence and machine learning have a role to play in power systems going forward. In particular, it may prove instrumental in the management of risks such as those which manifested on 28 April, as well as to assist with the transition to clean energy.